CONVERSION PROSPECTING FOR DYNAMIC CONTENT RECOMMENDATION

§101 §103

DETAILED ACTION Continued Examination Under 37 CFR 1.114 A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection on June 22, 2026. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on May 25, 2026 has been entered. Claims 1, 12, and 16 have been amended. Claims 1-20 are presented for examination. Notice of Pre-AIA or AIA Status The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . Response to Arguments Applicant's arguments filed May 25, 2026 have been fully considered but they are not persuasive. Regarding the rejection under 35 U.S.C. § 101, Applicant asserts that “the present claims provide for improvements to the technical field of computer-implemented content serving and recommendation systems, and are indicative of integration into a practical application (see MPEP 2106.05(a)).” (Page 10 of Applicant’s response) Regarding the performance of an internal preliminary auction to decrease load on a serving system of the content serving platform (recited in claims 1 and 16), the decrease in load is a natural result of filtering data content so that a smaller amount of data needs to be processed. Both a human or machine would generally have a processing and/or computational load decreased when less data needs to be processed. The claims do not present a specific technical manner in which a machine would particularly be able to decrease a load. Processing a smaller amount of data alone to decrease load without any specific technical details is not deemed to present a technical solution specific to a technical problem. Furthermore, stating that only a certain number of products continue to the next stage of processing is simply another step of filtering content. Regarding the prior art rejections, Applicant submits that the cited references do not address the claims as currently amended. The Lv, Kaufman, and Dirac references have been introduced into the rejections in order to help address the claim amendments. Claim Rejections - 35 USC § 101 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claimed invention is directed to “providing conversion prospecting for dynamic content recommendation” (Spec: ¶ 2) without significantly more. Step Analysis 1: Statutory Category? Yes – The claims fall within at least one of the four categories of patent eligible subject matter. Process (claims 1-11), Apparatus (claims 16-20), Article of Manufacture (claims 12-15) Independent claims: Step Analysis 2A – Prong 1: Judicial Exception Recited? Yes – Aside from the additional elements identified in Step 2A – Prong 2 below, the claims recite: [Claims 1, 12, 16] A method comprising: forming a model to predict conversion-given-click probabilities of users and products, where the forming comprises incrementally forming the model on (i) sets of positive events corresponding to post-click conversions occurring after a user clicks on a content with recommending at least one product and (ii) sets of negative events corresponding to post-click conversions not occurring after a user clicks on a content item recommending at least one product; publishing the model for use in a content serving flow, wherein, due to serving limitations, the published model is bounded to K products selected to include products having higher amounts of conversions than other products and that exceed a predefined minimum number of conversions [claims 1, 12 only]; generating, utilizing the published model, a prediction of a conversion-given-click probability that a user will perform an action in relation to a product [“published model” for claims 1, 12 only; “model” for Claim 16]; generating a bid for the user and the product based upon a selection of a lesser of (i) a value based upon a combination of the conversion-given-click probability that the user will perform the action in relation to the product with a target cost per action corresponding to a performance goal of a content provider and (ii) a content provider bid for the product, wherein the selection of the lesser enforces a performance constraint derived from the target cost per action and a spending constraint defined by the content provider bid such that the bid satisfies both the performance constraint and the spending constraint; determining whether the product is eligible to compete in an auction hosted by the content serving platform by comparing the bid to a threshold condition, wherein the product is determined to be eligible based upon the bid satisfying the threshold condition; performing, to decrease load on a serving system, an internal preliminary auction among a plurality of eligible products using scores for the eligible products, wherein each score is based on the bid for the corresponding eligible product multiplied by a probability of a click generated by a click prediction model, and selecting a top-L subset of the eligible products to proceed to a next stage of processing by the content serving platform [claims 1, 16 only]; and transmitting content items of products, from the top-L subset, selected for display to users [“from the top-L subset” for claims 1, 16 only]. Aside from the additional elements, the aforementioned claim details exemplify the abstract idea(s) of a mental process (since the details include concepts performed in the human mind, including an observation, evaluation, judgment, and/or opinion). As explained in MPEP § 2106.04(a)(1)(III), “[t]he courts consider a mental process (thinking) that ‘can be performed in the human mind, or by a human using a pen and paper’ to be an abstract idea. CyberSource Corp. v. Retail Decisions, Inc., 654 F.3d 1366, 1372, 99 USPQ2d 1690, 1695 (Fed. Cir. 2011).” The limitations reproduced above, as drafted, are a process that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. That is, other than reciting the additional elements identified in Step 2A – Prong 2 below, nothing in the claim elements precludes the steps from practically being performed in the mind and/or by a human using a pen and paper. For example, but for the recitations of generic computer and other processing components (identified in Step 2A – Prong 2 below), the respectively recited steps/functions of the claims, as drafted and set forth above, are a process that, under its broadest reasonable interpretation, covers performance of the limitations in the mind and/or with the use of pen and paper. A human user can form a model, generate the recited information, make the recited determinations, perform the recited calculations, evaluate the recited mathematical relationships, set bids, make the recited determinations, evaluate the costs, transmit information for display, etc. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind (and/or with pen and paper) but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claims recite an abstract idea. Aside from the additional elements, the aforementioned claim details exemplify a method of organizing human activity (since the details include examples of commercial or legal interactions, including advertising, marketing or sales activities or behaviors, and/or business relations and managing personal behavior or relationships or interactions between people, including social activities, teaching, and following rules or instructions). More specifically, the evaluated process is related to “providing conversion prospecting for dynamic content recommendation” (Spec: ¶ 2), which (under its broadest reasonable interpretation) is an example of marketing and managing personal behavior (i.e., organizing human activity); therefore, aside from the recitations of generic computer and other processing components (identified in Step 2A – Prong 2 below), the limitations identified in the more detailed claim listing above encompass the abstract idea of organizing human activity. Publishing the trained model for use in a content serving flow, wherein, due to serving limitations, the published model is bounded to K products selected to include products having higher amounts of conversions than other products and that exceed a predefined minimum number of conversions [claims 1, 12 only] is an example of filtering content. Performing, to decrease load on a serving system, an internal preliminary auction among a plurality of eligible products using scores for the eligible products, wherein each score is based on the bid for the corresponding eligible product multiplied by a probability of a click generated by a click prediction model, and selecting a top-L subset of the eligible products to proceed to a next stage of processing by the content serving platform [claims 1, 16 only] is another example of filtering content. MPEP § 2106.04(a)(2)(II)(C) cites the following as an example of managing personal behavior, i.e., organizing human activity: “filtering content, BASCOM Global Internet v. AT&T Mobility, LLC, 827 F.3d 1341, 1345-46, 119 USPQ2d 1236, 1239 (Fed. Cir. 2016) (finding that filtering content was an abstract idea under step 2A, but reversing an invalidity judgment of ineligibility due to an inadequate step 2B analysis).” MPEP § 2106.04(a)(2)(III)(D) cites the following as an example of a mental process: “An application program interface for extracting and processing information from a diversity of types of hard copy documents – Content Extraction, 776 F.3d at 1345, 113 USPQ2d at 1356.” 2A – Prong 2: Integrated into a Practical Application? No – The judicial exception(s) is/are not integrated into a practical application. Claim 1 recites a method executing on a processor of a computing device that causes the computing device to perform the recited operations. Claim 1 also recites training a model to predict conversion-given-click probabilities of users and products, where the training comprises incrementally training the model on sets of positive events from a content serving platform and sets of negative events from a content serving platform, and transmitting content items of products selected by the content serving platform to devices for display to users. Claim 1 publishes the trained model for use in a content serving flow of the content serving platform and performs an internal preliminary auction to decrease load on a serving system of the content serving platform. Claim 12 recites a non-transitory machine readable medium having stored thereon processor-executable instructions that when executed cause performance of the recited operations. Claim 12 also recites training a model to predict conversion-given-click probabilities of users and products, where the training comprises incrementally training the model on sets of positive events from a content serving platform and sets of negative events from a content serving platform, and transmitting content items of products selected by the content serving platform to devices for display to users. Claim 12 publishes the trained model for use in a content serving flow of the content serving platform. Claim 16 recites a computing device comprising: a processor; and memory comprising processor-executable instructions that when executed by the processor cause performance of the recited operations. Claim 16 also recites training a model to predict conversion-given-click probabilities of users and products, where the training comprises incrementally training the model on sets of positive events from a content serving platform and sets of negative events from a content serving platform, and transmitting content items of products selected by the content serving platform to devices for display to users. Claim 16 performs an internal preliminary auction to decrease load on a serving system of the content serving platform. The claims as a whole merely describe how to generally “apply” the abstract idea(s) in a computer environment. The claimed processing elements are recited at a high level of generality and are merely invoked as a tool to perform the abstract idea(s). Simply implementing the abstract idea(s) on a general-purpose processor is not a practical application of the abstract idea(s); Applicant’s specification discloses that the invention may be implemented using general-purpose processing elements and other generic components (Spec: ¶¶ 17-31, 61). The use of a processor/processing elements (e.g., as recited in all of the claims) facilitates generic processor operations. The use of a memory or machine-readable media with executable instructions facilitates generic processor operations. The additional elements are recited at a high-level of generality (i.e., as generic processing elements performing generic computer functions) such that the incorporation of the additional processing elements amounts to no more than mere instructions to apply the judicial exception(s) using generic computer components. There is no indication in the Specification that the steps/functions of the claims require any inventive programming or necessitate any specialized or other inventive computer components (i.e., the steps/functions of the claims may be implemented using capabilities of general-purpose computer components). Accordingly, the additional elements do not integrate the abstract ideas into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claims are directed to an abstract idea(s). The processing components presented in the claims simply utilize the capabilities of a general-purpose computer and are, thus, merely tools to implement the abstract idea(s). As seen in MPEP § 2106.05(a)(I) and § 2106.05(f)(2), the court found that accelerating a process when the increased speed solely comes from the capabilities of a general-purpose computer is not sufficient to show an improvement in computer-functionality and it amounts to a mere invocation of computers or machinery as a tool to perform an existing process (see FairWarning IP, LLC v. Iatric Sys., 839 F.3d 1089, 1095, 120 USPQ2d 1293, 1296 (Fed. Cir. 2016)). Considering that the implementation of the machine learning model and/or the training of the model is performed using generic processing elements, such an implementation is presented as a generic recitation of machine learning in the claims and as a general link to technology. The machine learning-based processing elements are simply tools to generally automate the underlying process that could be performed by a human. It is further noted that, as described in Applicant’s Specification, the machine learning operations are generic machine learning operations (Spec: ¶¶ 35, 39, 44). The Specification presents no assertion that there is any improvement in the automated machine learning process itself. Such a generic recitation of machine learning, as recited in the claims, is little more than automating an analogous process that can be performed by a human. There is no transformation or reduction of a particular article to a different state or thing recited in the claims. Regarding the performance of an internal preliminary auction to decrease load on a serving system of the content serving platform (recited in claims 1 and 16), the decrease in load is a natural result of filtering data content so that a smaller amount of data needs to be processed. Both a human or machine would generally have a processing and/or computational load decreased when less data needs to be processed. The claims do not present a specific technical manner in which a machine would particularly be able to decrease a load. Processing a smaller amount of data alone to decrease load without any specific technical details is not deemed to present a technical solution specific to a technical problem. 2B: Claim(s) Provide(s) an Inventive Concept? No – The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception(s). As discussed above with respect to integration of the abstract idea(s) into a practical application, the use of the additional elements to perform the steps identified in Step 2A – Prong 1 above amounts to no more than mere instructions to apply the exceptions using a generic computer component(s). Mere instructions to apply an exception using a generic computer component(s) cannot provide an inventive concept. The claims are not patent eligible. Dependent claims: Step Analysis 2A – Prong 1: Judicial Exception Recited? Yes – Aside from the additional elements identified in Step 2A – Prong 2 below, the claims recite: [Claim 2] wherein a positive event corresponds to a first user performing the action after interacting with a first content item, and wherein a negative event corresponds to the first user interacting with a second content item without subsequently performing the action. [Claim 3] wherein the model represents products with product features that include product identifiers, product set identifiers, and content provider identifiers. [Claim 4] wherein the model comprises a multi-value feature containing a list of content item campaigns with a highest click through rate of a first user during a time period. [Claim 5] wherein the model comprises a content serving platform experiment identifier. [Claim 6] wherein the model comprises a page section corresponding to a webpage identifier where a user viewed a content item. [Claim 7] adjusting, during forming of the model, predictions generated by the model based upon the model under-predicting due to the model being formed on an additional negative event for each positive event. [Claim 8] calculating the target cost per action by applying a factor to an average spend divided by a number of conversions of products of a content provider of the product, wherein the products are part of retargeting impressions. [Claim 9] setting the bid to guarantee that the user is eligible for the product if an expected value satisfies a content provider target. [Claim 10] setting the bid to guarantee that a value of the bid does not exceed an amount the content provider will pay for a click. [Claim 11] determining that the user is eligible for the product based upon the bid exceeding a floor price. [Claim 13] bounding the model by a set number of products selected to include products having higher amounts of conversions than other products and that exceed a predefined minimum number of conversions. [Claim 14] reducing the target cost per action to reduce a value of the bid and to decrease a pool of eligible users for the product. [Claim 15] increasing the target cost per action to increase the value of the bid and to increase a pool of eligible users for the product. [Claim 17] bounding the model by a set number of products selected to include products having higher amounts of conversions than other products and that exceed a predefined minimum number of conversions. [Claim 18] reducing the target cost per action to reduce a value of the bid and to decrease a pool of eligible users for the product. [Claim 19] increasing the target cost per action to increase a value of the bid and to increase a pool of eligible users for the product. [Claim 20] calculating the target cost per action by applying a factor to an average spend divided by a number of conversions of products of a content provider of the product. The dependent claims further recite details of the abstract ideas identified in the independent claims above. Aside from the additional elements, the aforementioned claim details exemplify the abstract idea(s) of a mental process (since the details include concepts performed in the human mind, including an observation, evaluation, judgment, and/or opinion). As explained in MPEP § 2106.04(a)(1)(III), “[t]he courts consider a mental process (thinking) that ‘can be performed in the human mind, or by a human using a pen and paper’ to be an abstract idea. CyberSource Corp. v. Retail Decisions, Inc., 654 F.3d 1366, 1372, 99 USPQ2d 1690, 1695 (Fed. Cir. 2011).” The limitations reproduced above, as drafted, are a process that, under its broadest reasonable interpretation, covers performance of the limitations in the mind but for the recitation of generic computer components. That is, other than reciting the additional elements identified in Step 2A – Prong 2 below, nothing in the claim elements precludes the steps from practically being performed in the mind and/or by a human using a pen and paper. For example, but for the recitations of generic computer and other processing components (identified in Step 2A – Prong 2 below), the respectively recited steps/functions of the claims, as drafted and set forth above, are a process that, under its broadest reasonable interpretation, covers performance of the limitations in the mind and/or with the use of pen and paper. A human user can form a model, generate the recited information, make the recited determinations, perform the recited calculations, evaluate the recited mathematical relationships, set bids, make the recited determinations, evaluate the costs, transmit information for display, etc. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind (and/or with pen and paper) but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claims recite an abstract idea. Aside from the additional elements, the aforementioned claim details exemplify a method of organizing human activity (since the details include examples of commercial or legal interactions, including advertising, marketing or sales activities or behaviors, and/or business relations and managing personal behavior or relationships or interactions between people, including social activities, teaching, and following rules or instructions). More specifically, the evaluated process is related to “providing conversion prospecting for dynamic content recommendation” (Spec: ¶ 2), which (under its broadest reasonable interpretation) is an example of marketing and managing personal behavior (i.e., organizing human activity); therefore, aside from the recitations of generic computer and other processing components (identified in Step 2A – Prong 2 below), the limitations identified in the more detailed claim listing above encompass the abstract idea of organizing human activity. Publishing the trained model for use in a content serving flow, wherein, due to serving limitations, the published model is bounded to K products selected to include products having higher amounts of conversions than other products and that exceed a predefined minimum number of conversions [claims 1, 12 only] is an example of filtering content. Performing, to decrease load on a serving system, an internal preliminary auction among a plurality of eligible products using scores for the eligible products, wherein each score is based on the bid for the corresponding eligible product multiplied by a probability of a click generated by a click prediction model, and selecting a top-L subset of the eligible products to proceed to a next stage of processing by the content serving platform [claims 1, 16 only] is another example of filtering content. MPEP § 2106.04(a)(2)(II)(C) cites the following as an example of managing personal behavior, i.e., organizing human activity: “filtering content, BASCOM Global Internet v. AT&T Mobility, LLC, 827 F.3d 1341, 1345-46, 119 USPQ2d 1236, 1239 (Fed. Cir. 2016) (finding that filtering content was an abstract idea under step 2A, but reversing an invalidity judgment of ineligibility due to an inadequate step 2B analysis).” MPEP § 2106.04(a)(2)(III)(D) cites the following as an example of a mental process: “An application program interface for extracting and processing information from a diversity of types of hard copy documents – Content Extraction, 776 F.3d at 1345, 113 USPQ2d at 1356.” Various calculating steps are recited throughout claims 8 and 20 and these are examples of mathematical concepts. 2A – Prong 2: Integrated into a Practical Application? No – The judicial exception(s) is/are not integrated into a practical application. The dependent claims incorporate the additional elements of the independent claims from which they depend. Claims 1-11 recite a method executing on a processor of a computing device that causes the computing device to perform the recited operations. Claim 1 also recites training a model to predict conversion-given-click probabilities of users and products, where the training comprises incrementally training the model on sets of positive events and sets of negative events from a content serving platform, and transmitting content items of products selected by the content serving platform to devices for display to users. Claim 1 publishes the trained model for use in a content serving flow of the content serving platform and performs an internal preliminary auction to decrease load on a serving system of the content serving platform. Claim 7 recites adjusting, during training of the model, predictions generated by the model based upon the model under-predicting due to the model being trained on an additional negative event for each positive event. Claims 12-15 recites a non-transitory machine readable medium having stored thereon processor-executable instructions that when executed cause performance of the recited operations. Claim 12 also recites training a model to predict conversion-given-click probabilities of users and products, where the training comprises incrementally training the model on sets of positive events and sets of negative events from a content serving platform, and transmitting content items of products selected by the content serving platform to devices for display to users. Claim 12 publishes the trained model for use in a content serving flow of the content serving platform. Claims 16-20 recites a computing device comprising: a processor; and memory comprising processor-executable instructions that when executed by the processor cause performance of the recited operations. Claim 16 also recites training a model to predict conversion-given-click probabilities of users and products, where the training comprises incrementally training the model on sets of positive events and sets of negative events from a content serving platform, and transmitting content items of products selected by the content serving platform to devices for display to users. Claim 16 performs an internal preliminary auction to decrease load on a serving system of the content serving platform. The claims as a whole merely describe how to generally “apply” the abstract idea(s) in a computer environment. The claimed processing elements are recited at a high level of generality and are merely invoked as a tool to perform the abstract idea(s). Simply implementing the abstract idea(s) on a general-purpose processor is not a practical application of the abstract idea(s); Applicant’s specification discloses that the invention may be implemented using general-purpose processing elements and other generic components (Spec: ¶¶ 17-31, 61). The use of a processor/processing elements (e.g., as recited in all of the claims) facilitates generic processor operations. The use of a memory or machine-readable media with executable instructions facilitates generic processor operations. The additional elements are recited at a high-level of generality (i.e., as generic processing elements performing generic computer functions) such that the incorporation of the additional processing elements amounts to no more than mere instructions to apply the judicial exception(s) using generic computer components. There is no indication in the Specification that the steps/functions of the claims require any inventive programming or necessitate any specialized or other inventive computer components (i.e., the steps/functions of the claims may be implemented using capabilities of general-purpose computer components). Accordingly, the additional elements do not integrate the abstract ideas into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claims are directed to an abstract idea(s). The processing components presented in the claims simply utilize the capabilities of a general-purpose computer and are, thus, merely tools to implement the abstract idea(s). As seen in MPEP § 2106.05(a)(I) and § 2106.05(f)(2), the court found that accelerating a process when the increased speed solely comes from the capabilities of a general-purpose computer is not sufficient to show an improvement in computer-functionality and it amounts to a mere invocation of computers or machinery as a tool to perform an existing process (see FairWarning IP, LLC v. Iatric Sys., 839 F.3d 1089, 1095, 120 USPQ2d 1293, 1296 (Fed. Cir. 2016)). Considering that the implementation of the machine learning model and/or the training of the model is performed using generic processing elements, such an implementation is presented as a generic recitation of machine learning in the claims and as a general link to technology. The machine learning-based processing elements are simply tools to generally automate the underlying process that could be performed by a human. It is further noted that, as described in Applicant’s Specification, the machine learning operations are generic machine learning operations (Spec: ¶¶ 35, 39, 44). The Specification presents no assertion that there is any improvement in the automated machine learning process itself. Such a generic recitation of machine learning, as recited in the claims, is little more than automating an analogous process that can be performed by a human. There is no transformation or reduction of a particular article to a different state or thing recited in the claims. Regarding the performance of an internal preliminary auction to decrease load on a serving system of the content serving platform (recited in claims 1 and 16), the decrease in load is a natural result of filtering data content so that a smaller amount of data needs to be processed. Both a human or machine would generally have a processing and/or computational load decreased when less data needs to be processed. The claims do not present a specific technical manner in which a machine would particularly be able to decrease a load. Processing a smaller amount of data alone to decrease load without any specific technical details is not deemed to present a technical solution specific to a technical problem. 2B: Claim(s) Provide(s) an Inventive Concept? No – The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception(s). As discussed above with respect to integration of the abstract idea(s) into a practical application, the use of the additional elements to perform the steps identified in Step 2A – Prong 1 above amounts to no more than mere instructions to apply the exceptions using a generic computer component(s). Mere instructions to apply an exception using a generic computer component(s) cannot provide an inventive concept. The claims are not patent eligible. Claim Rejections - 35 USC § 103 The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action: A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made. Claims 1-5 and 9-11 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (US 2008/0275757) in view of Lv et al. (CN-110570232-A, citing the English Translation) in view of Kaufman et al. (US 2024/0370733) in view of Dirac et al. (US 2015/0379430). [Claim 1] Sharma discloses a method executing on a processor of a computing device that causes the computing device to perform operations (¶¶ 92-103) comprising: training a model to predict conversion-given-click probabilities of users and products, where the training comprises incrementally training the model on sets of positive events, from a content serving platform, corresponding to post-click conversions occurring after a user clicks on a content item recommending at least one product and (ii) sets of negative events, from the content serving platform, corresponding to post-click conversions not occurring after a user clicks on a content item recommending at least one product (¶ 45 – “the pCTR [predicted clickthrough rate] may be derived by the learning model 202 using historical data (e.g., clickthrough data).”; ¶ 11 – “In some implementations, a correction factor may be computed to better predict the conversion rate. For example, the correction factor may be computed using an iterative process (e.g., by a learning model) that compensates for the deviation error of the predicted conversion rate within a bidding period. The iterative process may employ historical performance data to obtain an accurate estimated click-based bid. The correction factor may be automatically adjusted in an adaptive way to mitigate changes or fluctuations of the predicted conversion rate so as to yield an accurate estimated click-based bid as a function of the target bid.”; ¶ 63 – “ In some implementations, the correction factor γ in [11] may be computed (e.g., by the learning model 202) using an iterative process (e.g., a feedback loop) that compensates for the deviation error of the pCVR within a bidding period. The iterative process may employ historical performance data to obtain an accurate eCPC. The correction factor γ may be automatically adjusted in an adaptive way to mitigate changes or fluctuations of the predicted conversion rate so as to yield an accurate estimated CPC as a function of a CPA bid.”; ¶ 64 – “In some implementations, the iterations can be performed until a predetermined, dynamically determined, or other optimal value is reached for the correction factor. In certain implementations, a correction factor may be approximated (e.g., by the learning model 202) before iterations are performed, such as before a threshold number of iterations are performed. In other implementations, no further iteration is performed if the correction factor changes by less than a threshold amount after a particular iteration.”; ¶ 30 – “The advertisement data may include data relating to ads previously provided to users 218 and whether the ads were selected or not selected by the users 218.”; ¶ 33 – “Ads presented to the user, whether the user selected or did not select the ad(s), and the document(s) that the user accessed when presented with the ad(s) may be forwarded to the ad ranking model to create an empirical model so as to improve ad ranking.”; ¶ 52 – “In some implementations, the target bid specified by the advertisers 214 may be provided by the ad server 206 to the learning model 204 where the target bid can be combined with a predicted conversion rate to produce a new or adjusted maximum CPC bid. The learning model 202, for example, can be used to compute the pCVR [predicted conversion rate] for a potential ad impression by collecting the number of clicks and conversions for each impression context feature of interest. As discussed above, a conversion rate defines a ratio of the number of conversions (e.g., # of sales generated by a given ad) to the number of clicks of the ad (i.e., # of visits to advertiser's web property from the ad). Thus, statistics can be calculated based on these numbers for use in predicting a conversion rate. Once the pCVR is determined, this parameter may be used with the target bid (e.g., multiplied by the target bid) to automatically adjust the advertiser's default click-based bid (e.g., maximum CPC bid) or compute a new click-based bid.”; ¶ 40 – “A "conversion" is said to occur when a user consummates a transaction related to a previously served ad. What constitutes a conversion may vary from case to case and can be determined in a variety of ways. For example, it may be the case that a conversion occurs when a user clicks on an ad, is referred to the advertiser's web page, and consummates a purchase before leaving that web page. Alternatively, a conversion may be defined as a user being shown an ad, and making a purchase on the advertiser's web page within a predetermined time (e.g., seven days). Many other definitions of what constitutes a conversion are possible. For example, a conversion also may include signing up to be a member of a web site, filling out an online form, making a purchase upon contacting an advertiser via online creatives and the like.” A purchase made after clicking on an ad is an example of a post-click conversion, i.e., a positive event. A conversion rate may convey a number of sales results from a total number of clicks (¶ 41), which means that both positive events and negative events are used to evaluate post-click conversions. When a conversion is defined as the purchase, not making the purchase is an example of a negative event. The ads may be correlated with products (¶¶ 5, 34); therefore, a purchase of an advertised product is an example of a post-click conversion related to an ad, i.e., item content, recommending the at least one product, i.e., the product that is the subject of the ad. A correction factor is computed using an iterative process and is automatically adjusted adaptively as part of a machine learning process to adjust bids and predictions over time and in response to changing prediction conditions (¶¶ 11, 53-55, 61-64, 70). This information is used to rank ads for an ad campaign (¶ 42).); generating, utilizing the published (i.e., selected/applied) model, a prediction of a conversion-given-click probability that a user will perform an action in relation to a product (¶ 52 – “In some implementations, the target bid specified by the advertisers 214 may be provided by the ad server 206 to the learning model 204 where the target bid can be combined with a predicted conversion rate to produce a new or adjusted maximum CPC bid. The learning model 202, for example, can be used to compute the pCVR [predicted conversion rate] for a potential ad impression by collecting the number of clicks and conversions for each impression context feature of interest. As discussed above, a conversion rate defines a ratio of the number of conversions (e.g., # of sales generated by a given ad) to the number of clicks of the ad (i.e., # of visits to advertiser's web property from the ad). Thus, statistics can be calculated based on these numbers for use in predicting a conversion rate. Once the pCVR is determined, this parameter may be used with the target bid (e.g., multiplied by the target bid) to automatically adjust the advertiser's default click-based bid (e.g., maximum CPC bid) or compute a new click-based bid.”; ¶ 40 – “A "conversion" is said to occur when a user consummates a transaction related to a previously served ad. What constitutes a conversion may vary from case to case and can be determined in a variety of ways. For example, it may be the case that a conversion occurs when a user clicks on an ad, is referred to the advertiser's web page, and consummates a purchase before leaving that web page. Alternatively, a conversion may be defined as a user being shown an ad, and making a purchase on the advertiser's web page within a predetermined time (e.g., seven days). Many other definitions of what constitutes a conversion are possible. For example, a conversion also may include signing up to be a member of a web site, filling out an online form, making a purchase upon contacting an advertiser via online creatives and the like.”; The ads may be correlated with products (¶ 34). By nature of the association of the products with the ads, a ranking of the ads and conversion rates may be seen as also reflecting a ranking of the advertised products.); generating a bid for the user and the product based upon the conversion-given-click probability that the user will perform the action in relation to the product, a content provider bid for the product, and a target cost per action corresponding to a performance goal of a content provider (¶ 29 – “The learning model 202, which can be coupled to the ad server 206 and the conversion data repository 208, may include a statistical and probability model constructed using statistical techniques. Such techniques may include, for example, logistic regression, regression trees, boosted stumps, or any other statistical modeling techniques. In some implementations, the learning model 202 provides a predicted conversion rate ("pCVR"), which can be used to perform a metric conversion, as described below with reference to FIGS. 3 and 4.”; ¶ 53 – “In some implementations, the learning model 202 is a machine learning system model that includes rules for mapping impression context features to conversion rate predictions. The rules may include, for example, a probability multiplier for each context feature. For example, a user from USA may be assigned a probability multiplier of 0.85, and an ad appearing on a particular news web site may be assigned a probability multiplier of 1.1. To predict a conversion rate, the default conversion rate can be aggregated with the probability multiplier for each relevant feature. Using the above example, for an ad having a default conversion rate of 0.2% that is shown on the particular news web site to a user from USA, the predicted conversion rate for the ad would be 0.187% (0.2%*1.1*0.85).”; ¶ 13 – “In some implementations, a method includes: obtaining input specifying a first metric value associated with an advertisement; determining a predicted conversion rate for a potential impression of the advertisement; estimating a second metric value based on the first metric value and the predicted conversion rate; compensating based on the second metric value; and debiting based on the first metric value. The first metric value may be a value based on a Cost-Per-Action model, and the second metric value may be based on a Cost-Per-Click model. Alternatively, the first metric value may be a value based on one of Cost-Per-Click model or Cost-Per-Action model, and the second metric value may be based on a Cost-Per-Impression model.”; ¶ 35 – “Each ad or ad group may include individual price information (e.g., cost, average cost, or maximum cost (per impression, per selection, per conversion, etc.)). For example, the advertiser 102 may specify a maximum monetary value with the advertising management system 104 as to how much the advertiser 102 is willing to pay per user click, impression or conversion per ad or ad group. The maximum monetary value may be based on the number of impressions (e.g., CPM bidding), the number of clicks on an ad (e.g., CPC bidding), or the number of conversions (e.g., CPA bidding) generated in response to an ad. For example, if the advertiser 102 has selected a CPC bidding model, the advertiser 102 can enter a maximum CPC bid which represents the highest amount the advertiser 102 is willing to pay if an ad associated with an ad group receives a click. As another example, if the advertiser 102 has selected a CPA bidding model, the advertiser 102 can enter a maximum CPA bid which represents the highest amount an advertiser is willing to pay if an ad associated with that ad group generates a conversion. Based on the defined bidding model, the publishers 106 can be credited and the advertisers 102 can be debited accordingly.”; ¶ 43 – “Based on the ranking (and any adjustment thereof as a result of the weighted scores) determined during the auction, the identified ads can be selected for presentation.”; The ads may be correlated with products (¶ 34). By nature of the association of the products with the ads, a ranking of the ads and conversion rates may be seen as also reflecting a ranking of the advertised products and actions in relation to the advertised products.; ¶ 10 – “An advertiser specifies a target bid (e.g., CPA target bid or other target) for a conversion event associated with an ad. A predicted conversion rate or value is determined (e.g., empirically) for potential impressions of the ad based on conversion data (e.g., historical conversion data) for the ad and impression context data (e.g., current impression context data). The predicted conversion rate and target bid can be used to estimate a click-based bid. Publishers can be compensated based on the estimated click-based bid, while advertisers can be debited using the target bid originally specified.” Considering that an ad may have multiple impressions, this implies that the ads may be retargeted impressions. The ads may be correlated with products (¶ 34). By nature of the association of the products with the ads, a ranking of the ads and conversion rates may be seen as also reflecting a ranking of the advertised products and actions in relation to the advertised products.; ¶ 27 – “In some implementations, when the user 218 clicks an ad served by the ad server 206, the user 218 is directed to a landing page on the web property (e.g., a web site) of the advertiser 214. The user 218 may then perform a conversion event at the website (e.g., make a purchase, register). The conversion event generates conversion data which is sent to the system 200 and stored in a repository (e.g., MySQL® database). In this manner, a conversion history can be accumulated and maintained for each ad or ad group in an advertiser's ad campaign.”).; ¶ 40 – “A "conversion" is said to occur when a user consummates a transaction related to a previously served ad. What constitutes a conversion may vary from case to case and can be determined in a variety of ways. For example, it may be the case that a conversion occurs when a user clicks on an ad, is referred to the advertiser's web page, and consummates a purchase before leaving that web page. Alternatively, a conversion may be defined as a user being shown an ad, and making a purchase on the advertiser's web page within a predetermined time (e.g., seven days). Many other definitions of what constitutes a conversion are possible. For example, a conversion also may include signing up to be a member of a web site, filling out an online form, making a purchase upon contacting an advertiser via online creatives and the like.” A purchase made after clicking on an ad is an example of a post-click conversion, i.e., a positive event. A conversion rate may convey a number of sales results from a total number of clicks (¶ 41), which means that both positive events and negative events are used to evaluate post-click conversions. When a conversion is defined as the purchase, not making the purchase is an example of a negative event. The ads may be correlated with products (¶¶ 5, 34); therefore, a purchase of an advertised product is an example of a post-click conversion related to an ad, i.e., item content, recommending the at least one product, i.e., the product that is the subject of the ad. A correction factor is computed using an iterative process and is automatically adjusted adaptively as part of a machine learning process to adjust bids and predictions over time and in response to changing prediction conditions (¶¶ 11, 53-55, 61-64, 70). This information is used to rank ads for an ad campaign (¶ 42).); determining whether the product is eligible to compete in an auction hosted by the content serving platform by comparing the bid to a threshold condition, wherein the product is determined to be eligible based upon the bid satisfying the threshold condition (¶ 43 – “Based on the ranking (and any adjustment thereof as a result of the weighted scores) determined during the auction, the identified ads can be selected for presentation.”; ¶ 38 – “As another example, auction factors such as clickthrough rates (CTRs) and conversion rates (CVRs) can be ranked from largest to smallest.” The ads may be correlated with products (¶ 34). By nature of the association of the products with the ads, a ranking of the ads and conversion rates may be seen as also reflecting a ranking of the advertised products.; ¶ 32 – “The rank of an ad may, in some cases, correspond to the probability that a user will select the ad when accessing a particular document. Each ad can then be served to the user based on a respective rank. For example, the top one or more ads may be served to the users 218. Alternatively, ads that have ranks above a predetermined threshold may be served to the users 218.”; ¶ 36 – “When an ad request is received, an ad that corresponds to the received ad request is identified. If more than one ad has been identified, an auction can be conducted to identify which ad to be served. During the auction, the ads may be ranked in accordance with one or more associated ad campaign parameters. The one or more ad campaign parameters may include, without limitation, default bids (e.g., CPC, CPA or CPM bids), daily budget defined by the advertisers 102 (e.g., at the time of registering an ad campaign), and relevance of an ad that can be determined by various methods, such as by inferring a high relevance for an ad with respect to, for example, a particular keyword query.”); and transmitting content items of products selected by the content serving platform to devices for display to users (¶ 26 – “In some implementations, a publisher 216 can request an ad from the ad server 206. In response to the request, one or more ads (e.g., image ads) are sent to the publisher 216. The ad(s) can be placed on, for example, a web property owned or operated by the publisher 216 (e.g., a web site). In some implementations, the web page can have a page content identifier (ID), which can be used by the ad server 206 to determine ad context for targeting ads. These implementations target ads in the hope that the users, e.g., user 218, will be more receptive to targeted ads than to untargeted ads.”; ¶ 27 – “In some implementations, when the user 218 clicks an ad served by the ad server 206, the user 218 is directed to a landing page on the web property (e.g., a web site) of the advertiser 214. The user 218 may then perform a conversion event at the website (e.g., make a purchase, register).” It is understood that the user would need a display device to view content on a website.; The ads may be correlated with products (¶ 34).). Sharma does not explicitly disclose publishing the trained model for use in a content serving flow of the content serving platform, wherein, due to serving limitations, the published model is bounded to K products selected to include products having higher amounts of conversions than other products and that exceed a predefined minimum number of conversions. In the field of advertising and conversion rate prediction models, Lv discloses that a machine learning model training the conversion model may be selected based on a sample size compared to various quantitative ranges of sample sizes. On page 12, Lv explicitly states: In the embodiment of the invention, can select different learning model according to different magnitude. For example, equivalent level is large, it can choose target learning model is deep learning model, the equivalent level is small, can select the target learning model is a machine learning model. For example, because of different advertisement transformed sample difference is large, the advertisement may have a million converted samples, some advertisement only with hundreds of transformed samples. mode, the conversion rate prediction model embodiment of the invention uses hierarchical training. sample magnitude according to the machine learning model, to classifying and training the conversion model according to the following rule: I, when the user behaviour sample number is greater than 10000, optionally deep learning model training estimating model conversion rate, deep learning model of use include, but are not limited to, deep neural network (Deep confirm Topology Control algorithm for Network (DNN) model, a convolutional neural network (Convolutional confirm Topology Control algorithm for Networks, CNN) model, the depth extent (wide&deep) model, a depth factor decomposer (brachioradialis Factorization Machine. deepFM) model, and the like. II, when user behaviour sample number greater than 1000 and less than 10000, selecting the machine learning model training the conversion model, the machine learning model including but not limited to logistic regression (logisitic describing, LR) model, a gradient boosted trees (zeroth gradient boosting decision tree, GBDT) model, a support vector machine (SupportVector Machine, SVM) model, a random forest model and so on. III, when the user behaviour sample number is less than 1000, using statistical method, directly counting the global conversion of the advertisement, the conversion rate of each advertisement material and conversion rate of each media. For the conversion number less than 50 of advertisement materials and media, it is considered that statistical confidence is too low, replaced by a global conversion conversion rate of the material and media for subsequent calculation. The embodiment of the invention can be adjusted according to the magnitude of the user behaviour sample number corresponding to the selected learning model, selecting proper learning model for training according to the magnitude of the sample number, the estimated training effect of the model conversion. The Examiner submits that it would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s invention to modify Sharma to perform the step of publishing the trained model for use in a content serving flow of the content serving platform, wherein, due to serving limitations, the published model is bounded to K products selected to include products having higher amounts of conversions than other products and that exceed a predefined minimum number of conversions in order to facilitate the selection of a model-sample size combination that “can improve the optimization efficiency of the advertisement effect” (as suggested in the Abstract of Lv). As discussed above, Sharma addresses the step of generating a bid for the user and the product based upon the conversion-given-click probability that the user will perform the action in relation to the product, a content provider bid for the product, and a target cost per action corresponding to a performance goal of a content provider. Sharma does not explicitly disclose: generating a bid for the user and the product based upon a selection of a lesser of (i) a value based upon a combination of the conversion-given-click probability that the user will perform the action in relation to the product with a target cost per action corresponding to a performance goal of a content provider and (ii) a content provider bid for the product, wherein the selection of the lesser enforces a performance constraint derived from the target cost per action and a spending constraint defined by the content provider bid such that the bid satisfies both the performance constraint and the spending constraint. Kaufman takes into account a budget available to the advertiser and a ratio of conversions to ad spend, and trains computer models based on a number of impressions that an advertiser plans to display (Kaufman: ¶¶ 65, 82, 83). Kaufman also evaluates trade-offs between advertising scale and precision and sets lowest and highest bid prices and conversion combinations that are acceptable to an advertiser, as seen in ¶¶ 102-103: [0102] Since targeted advertising involves a trade-off between precision (high conversion rate) and scale (number of impressions), the ability to make use of all inventory expands the available scale and provides a significant performance advantage. In general, expanding the available inventory allows the advertisers to either increase scale at a fixed precision, or increase precision at fixed scale. The base bid can be chosen to achieve a desired price per conversion or can be adjusted to achieve the desired scale. For example, the advertiser may start with a base bid corresponding the highest acceptable cost per conversion, and gradually lower the base bid to the lowest price (and correspondingly the lowest cost per conversion) before the campaign fails to deliver the desired scale. Conversely, the advertiser may start with a low base bid and increase it until it either reaches the desired scale or the maximum acceptable cost per conversion [0103] Other functions for setting the bid factor based on the probability of conversion may be used as well. For example, if the advertiser wishes to bid less often on inventory with a low conversion rate, a conversion rate threshold may be set, and inventory with conversion rate at the threshold may be set to have a bid factor of 1. The bid factor for inventory with greater conversion rates than the threshold can increase linearly with conversion rate, and the bid factor for inventory with lower conversion rates can decrease sublinearly with conversion rate. This way, inventory with low conversion rates is only used when available at a very low price resulting in a low cost per conversion. Determination of the bid factor may also include information not provided by the model. For example, if data on the win rate of auctions for a website is available, the bid factor can proportional to conversion rate, with a maximum bid price corresponding to a win rate threshold, a price at which a high percent of auctions will be won. The Examiner submits that it would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s invention to modify Sharma to perform the step of generating a bid for the user and the product based upon a selection of a lesser of (i) a value based upon a combination of the conversion-given-click probability that the user will perform the action in relation to the product with a target cost per action corresponding to a performance goal of a content provider and (ii) a content provider bid for the product, wherein the selection of the lesser enforces a performance constraint derived from the target cost per action and a spending constraint defined by the content provider bid such that the bid satisfies both the performance constraint and the spending constraint in order to increase the likelihood that desired conversion levels may be reached more cost-effectively and within an advertiser’s budgetary constraints (which Kaufman suggests by explaining problems that could otherwise arise, as seen in ¶¶ 4, 86 of Kaufman). As discussed above, Sharma generates, utilizing the published (i.e., selected/applied) model, a prediction of a conversion-given-click probability that a user will perform an action in relation to a product, determines whether the product is eligible to compete in an auction hosted by the content serving platform by comparing the bid to a threshold condition, wherein the product is determined to be eligible based upon the bid satisfying the threshold condition, and transmits content items of products selected by the content serving platform to devices for display to users. Also, Sharma determines a maximum adjusted bid based on a bid for an advertisement multiplied by a predicted conversion rate, as seen in ¶¶ 52-54: PNG media_image1.png 456 268 media_image1.png Greyscale In other words, Sharma addresses the claim limitation “wherein each score is based on the bid for the corresponding eligible product multiplied by a probability of a click generated by a click prediction model.” Sharma also maps the conversion rate predictions to impression context features (Sharma: ¶ 53 – “In some implementations, the learning model 202 is a machine learning system model that includes rules for mapping impression context features to conversion rate predictions. The rules may include, for example, a probability multiplier for each context feature. For example, a user from USA may be assigned a probability multiplier of 0.85, and an ad appearing on a particular news web site may be assigned a probability multiplier of 1.1. To predict a conversion rate, the default conversion rate can be aggregated with the probability multiplier for each relevant feature. Using the above example, for an ad having a default conversion rate of 0.2% that is shown on the particular news web site to a user from USA, the predicted conversion rate for the ad would be 0.187% (0.2%*1.1*0.85).”). Many features related to advertising may be used to train the learning model (Sharma: ¶ 30). Sharma does not explicitly disclose: performing, to decrease load on a serving system of the content serving platform, an internal preliminary auction among a plurality of eligible products using scores for the eligible products, wherein each score is based on the bid for the corresponding eligible product multiplied by a probability of a click generated by a click prediction model, and selecting a top-L subset of the eligible products to proceed to a next stage of processing by the content serving platform; and transmitting content items of products selected, from the top-L subset, by the content serving platform to devices for display to users. Given that Sharma’s trained features are related to modeling an auction for delivering content and features include related information, including conversion rate prediction information, and Sharma’s modeled information incorporates a score that is based on the bid for the corresponding eligible product multiplied by a probability of a click generated by a click prediction model, Sharma lends itself to improvements in the area of machine learning, including as they relate to selecting the most relevant information for making auction-related predictions. Dirac breaks down the training of a model into multiple stages and reduces the feature set size in subsequent iterations and also allows for the least relevant features to be eliminated and the most relevant features to be reintroduced, as needed in light of each particular learning iteration. For example, Dirac explains: [0273] After some subset of the features have been identified as pruning candidates and removed, as indicated by the arrow labeled 5255 in FIG. 52, the pruned feature set (comprising adjusted parameters 5222K* for features 5212K*) may no longer violate the TFSS constraint. In at least some embodiments, a sufficiently large fraction of the feature set may be pruned that additional features may again be added in one or more subsequent learning iterations, such as learning iteration 5210L shown in FIG. 52. Thus, the feature set size may grow again after being reduced via pruning Additional pruning may be required if the feature set size again exceeds TFSS eventually, and more features may be added after the additional pruning is completed. Any feature may be added to the feature set in a given learning iteration, including for example some of the features that were pruned earlier. By not restricting the set of features that can be considered for inclusion in the feature set during any learning iteration, and eliminating only those features at each pruning stage that are currently identified as contributing the least to the model's predictions, the technique illustrated in FIG. 52 may converge on a feature set that provides highly accurate predictions while limiting memory use during training. In addition, the reduction in the feature set size may also reduce the time it takes to load and execute the model during prediction runs—thus, the benefits of the technique may be obtained both during the training phase and in post-training-phase prediction runs. The Examiner submits that it would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s invention to modify Sharma to perform the steps of: performing, to decrease load on a serving system of the content serving platform, an internal preliminary auction among a plurality of eligible products using scores for the eligible products, wherein each score is based on the bid for the corresponding eligible product multiplied by a probability of a click generated by a click prediction model, and selecting a top-L subset of the eligible products to proceed to a next stage of processing by the content serving platform; and transmitting content items of products selected, from the top-L subset, by the content serving platform to devices for display to users because “[b]y not restricting the set of features that can be considered for inclusion in the feature set during any learning iteration, and eliminating only those features at each pruning stage that are currently identified as contributing the least to the model's predictions, the technique illustrated in FIG. 52 may converge on a feature set that provides highly accurate predictions while limiting memory use during training. In addition, the reduction in the feature set size may also reduce the time it takes to load and execute the model during prediction runs—thus, the benefits of the technique may be obtained both during the training phase and in post-training-phase prediction runs” (Dirac: ¶ 273) while presenting content most likely to be of interest to users. [Claim 2] Sharma discloses wherein a positive event corresponds to a first user performing the action after interacting with a first content item, and wherein a negative event corresponds to the first user interacting with a second content item without subsequently performing the action (¶ 30 – “In some implementations, the conversion data repository 208 may include one or more logical or physical memory devices configured to store a large data set (e.g., millions of instances and hundreds of thousands of features) that may be used, for example, to create and train the learning model 202. The data may include conversion data, advertisement information, such as advertisement data, user information, and document or content information, that may be used to create a model that may be used to determine a metric conversion rate. The advertisement data may include data relating to ads previously provided to users 218 and whether the ads were selected or not selected by the users 218.”). [Claim 3] Sharma discloses wherein the model represents products with product features that include product identifiers, product set identifiers, and content provider identifiers (¶ 34 – “Referring back to FIG. 1, each advertiser 102 may establish an advertising program with the advertising management system 104. An advertising program may include, for example, ad campaigns, creatives, targeting and the like. The advertiser 102 may define an "ad campaign" which can include one or more ad groups each including one or more ads. An ad group may define, for example, a product type (e.g., Hats or Pants), and a creative may include an ad textually or graphically defining the product type Each ad group or ad may include a start date, an end date, budget information, geographical targeting information, and syndication information.”). [Claim 4] Sharma discloses wherein the model comprises a multi-value feature containing a list of content item campaigns with a highest click through rate of a first user during a time period (¶ 42 – “An ad campaign parameter or auction factor also can be weighted to increase or decrease the impact that auction factor has on ad ranking. For example, an ad that has a higher CTR can be ranked above an ad that has a lower CTR, even if the ad with the lower CTR has an equal or greater default bid.”; ¶ 36 – “When an ad request is received, an ad that corresponds to the received ad request is identified. If more than one ad has been identified, an auction can be conducted to identify which ad to be served. During the auction, the ads may be ranked in accordance with one or more associated ad campaign parameters. The one or more ad campaign parameters may include, without limitation, default bids (e.g., CPC, CPA or CPM bids), daily budget defined by the advertisers 102 (e.g., at the time of registering an ad campaign), and relevance of an ad that can be determined by various methods, such as by inferring a high relevance for an ad with respect to, for example, a particular keyword query.”; ¶ 39 – “The CTR of an ad can be calculated by dividing the number of clickthroughs associated with the ad by the number of impressions of that ad over a given time period.”; ¶ 40 – “Alternatively, a conversion may be defined as a user being shown an ad, and making a purchase on the advertiser's web page within a predetermined time (e.g., seven days). Many other definitions of what constitutes a conversion are possible.”). [Claim 5] Sharma discloses wherein the model comprises a content serving platform experiment identifier (¶ 26 – “In some implementations, the web page can have a page content identifier (ID), which can be used by the ad server 206 to determine ad context for targeting ads. These implementations target ads in the hope that the users, e.g., user 218, will be more receptive to targeted ads than to untargeted ads.”; ¶ 98 – “The advertising management module 518 includes an ad server 520 and a web server 522. The advertising management module 518 further includes a learning model 524. The learning model 524 may perform and behave in a manner similar to learning model 202. The ad server 520 can be a server process or dedicated machine that is responsible for serving ads to publisher web properties and for tracking various information related to the ad placement (e.g., cookies, user URLs, page content, geographic information). The web server 522 (e.g., Apache web page server) serves web pages to advertisers and publishers and provides a means for advertisers and publishers to specify a target cost-per-action for use by the learning model 524 to dynamically compute or adjust an advertiser's click-based bid (e.g., Max CPC bid) or other performance metric.”). [Claim 9] Sharma discloses setting the bid to guarantee that the user is eligible for the product if an expected value satisfies a content provider target (¶ 77 – “ As shown, the process 300 begins with obtaining a target CPA bid and conversion rate (302). The target CPA bid may be a maximum CPA bid specified by the advertisers. Alternatively, a recommended CPA bid may be provided by the advertising management system 104 and used as the target CPA bid. In some implementations, the target CPA bid may be an average CPA bid specified by the advertisers. In another implementation, the target CPA bid may be a minimum CPA bid specified by the advertisers.”; ¶ 89 – “Process 400 then proceeds with predicting a conversion rate using the conversion data (410). In some implementations, the conversion data may include data associated with a default CPC bid and a target CPA bid specified by either the advertisers or the advertising management system. In these implementations, the conversion rate can be estimated by dividing a default CPC bid by a target CPA bid. Alternatively, the default maximum CPC bid can be used as default instead of predicting a conversion rate.”). [Claim 10] Sharma discloses setting the bid to guarantee that a value of the bid does not exceed an amount the content provider will pay for a click (¶ 35 – “Each ad or ad group may include individual price information (e.g., cost, average cost, or maximum cost (per impression, per selection, per conversion, etc.)). For example, the advertiser 102 may specify a maximum monetary value with the advertising management system 104 as to how much the advertiser 102 is willing to pay per user click, impression or conversion per ad or ad group.”). [Claim 11] Sharma discloses determining that the user is eligible for the product based upon the bid exceeding a floor price (¶ 77 – “ As shown, the process 300 begins with obtaining a target CPA bid and conversion rate (302). The target CPA bid may be a maximum CPA bid specified by the advertisers. Alternatively, a recommended CPA bid may be provided by the advertising management system 104 and used as the target CPA bid. In some implementations, the target CPA bid may be an average CPA bid specified by the advertisers. In another implementation, the target CPA bid may be a minimum CPA bid specified by the advertisers.”; The ads may be correlated with products (¶ 34). By nature of the association of the products with the ads, a ranking of the ads and conversion rates may be seen as also reflecting a ranking of the advertised products.). Claims 12-13 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (US 2008/0275757) in view of Lv et al. (CN-110570232-A, citing the English Translation) in view of Kaufman et al. (US 2024/0370733). [Claim 12] Claim 12 recites limitations already addressed by the rejection of claim 1 above; therefore, the same rejection applies. Furthermore, Sharma discloses a non-transitory machine readable medium having stored thereon processor-executable instructions that when executed cause performance of the disclosed operations (¶¶ 92-103). It is noted that claim 12 does not incorporate the limitation “performing, to decrease load on a serving system of the content serving platform, an internal preliminary auction among a plurality of eligible products using scores for the eligible products, wherein each score is based on the bid for the corresponding eligible product multiplied by a probability of a click generated by a click prediction model, and selecting a top-L subset of the eligible products to proceed to a next stage of processing by the content serving platform” recited in claim 1 and more specifically addressed, at least in the part, by the Dirac reference; therefore, (unlike in claim 1) the Dirac reference is not relied upon in the rejection of claim 12. [Claim 13] Sharma discloses bounding the model by a set number of products selected to include products having higher amounts of conversions than other products and that exceed a predefined minimum number of conversions (¶ 32 – “The ad ranking model may be used as part of a function to determine what ads to provide to users 218 when the users are accessing the documents. To facilitate data generation for use by the ad ranking model, information concerning a user and documents accessed by the user can be collected. As discussed above, information concerning a user may include IP address, cookie information, languages, geographical information and document information may include information relating to the documents accessed by the user (e.g., a URL of a web site visited by the user). Ads stored in the ad repository 210 may then be ranked based, at least in part, on the data stored by the learning model 202. The rank of an ad may, in some cases, correspond to the probability that a user will select the ad when accessing a particular document. Each ad can then be served to the user based on a respective rank. For example, the top one or more ads may be served to the users 218. Alternatively, ads that have ranks above a predetermined threshold may be served to the users 218.”; ¶ 34 – “The advertiser 102 may define an "ad campaign" which can include one or more ad groups each including one or more ads. An ad group may define, for example, a product type (e.g., Hats or Pants), and a creative may include an ad textually or graphically defining the product type Each ad group or ad may include a start date, an end date, budget information, geographical targeting information, and syndication information.”; ¶ 38 – “As another example, auction factors such as clickthrough rates (CTRs) and conversion rates (CVRs) can be ranked from largest to smallest.” The ads may be correlated with products. By nature of the association of the products with the ads, a ranking of the ads and conversion rates may be seen as also reflecting a ranking of the advertised products. A minimum threshold is an example of a lower bound and a largest/highest ranking is an example of an upper bound.). Claims 16-17 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (US 2008/0275757) in view of Kaufman et al. (US 2024/0370733) in view of Dirac et al. (US 2015/0379430). [Claim 16] Claim 16 recites limitations already addressed by the rejection of claim 1 above; therefore, the same rejection applies. Furthermore, Sharma discloses a computing device comprising: a processor (¶¶ 92-103); and memory comprising processor-executable instructions that when executed by the processor cause performance of the disclosed operations (¶¶ 92-103). It is noted that claim 16 does not incorporate the limitation “publishing the trained model for use in a content serving flow of the content serving platform, wherein, due to serving limitations, the published model is bounded to K products selected to include products having higher amounts of conversions than other products and that exceed a predefined minimum number of conversions” recited in claim 1 and more specifically addressed, at least in the part, by the Lv reference; therefore, (unlike in claim 1) the Lv reference is not relied upon in the rejection of claim 16. [Claim 17] Sharma discloses wherein the operations comprise: bounding the model by a set number of products selected to include products having higher amounts of conversions than other products and that exceed a predefined minimum number of conversions (¶ 32 – “The ad ranking model may be used as part of a function to determine what ads to provide to users 218 when the users are accessing the documents. To facilitate data generation for use by the ad ranking model, information concerning a user and documents accessed by the user can be collected. As discussed above, information concerning a user may include IP address, cookie information, languages, geographical information and document information may include information relating to the documents accessed by the user (e.g., a URL of a web site visited by the user). Ads stored in the ad repository 210 may then be ranked based, at least in part, on the data stored by the learning model 202. The rank of an ad may, in some cases, correspond to the probability that a user will select the ad when accessing a particular document. Each ad can then be served to the user based on a respective rank. For example, the top one or more ads may be served to the users 218. Alternatively, ads that have ranks above a predetermined threshold may be served to the users 218.”; ¶ 34 – “The advertiser 102 may define an "ad campaign" which can include one or more ad groups each including one or more ads. An ad group may define, for example, a product type (e.g., Hats or Pants), and a creative may include an ad textually or graphically defining the product type Each ad group or ad may include a start date, an end date, budget information, geographical targeting information, and syndication information.”; ¶ 38 – “As another example, auction factors such as clickthrough rates (CTRs) and conversion rates (CVRs) can be ranked from largest to smallest.” The ads may be correlated with products. By nature of the association of the products with the ads, a ranking of the ads and conversion rates may be seen as also reflecting a ranking of the advertised products. A minimum threshold is an example of a lower bound and a largest/highest ranking is an example of an upper bound.). Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (US 2008/0275757) in view of Lv et al. (CN-110570232-A, citing the English Translation) in view of Kaufman et al. (US 2024/0370733) in view of Dirac et al. (US 2015/0379430), as applied to claim 1 above, in view of Fernandez-Ruiz (US 2014/0372415). [Claim 6] Sharma discloses wherein the model comprises a page section corresponding to a webpage identifier (Sharma: ¶ 33 – “The position of the ads within a document accessed by the users 218 also may be based, at least in part, on the ranks of the ads. For example, higher ranking ads may be positioned at more prominent or visually-recognized locations then lower ranking ads. Ads presented to the user, whether the user selected or did not select the ad(s), and the document(s) that the user accessed when presented with the ad(s) may be forwarded to the ad ranking model to create an empirical model so as to improve ad ranking.”; ¶ 43 – “Based on the ranking (and any adjustment thereof as a result of the weighted scores) determined during the auction, the identified ads can be selected for presentation. For example, if the identified ads are to be presented in a web page that has four ad slots each slot displaying one ad, the four highest ranked ads can be selected for presentation. Further, the ranking established during the auction can be used to determine a display order. For example, the highest ranked ad can be assigned to the most prominent display position.”; ¶ 70 – “The bidding period "t" for which these parameters are measured can be empirically determined, or based on historical data.”; ¶ 40 – “A "conversion" is said to occur when a user consummates a transaction related to a previously served ad. What constitutes a conversion may vary from case to case and can be determined in a variety of ways. For example, it may be the case that a conversion occurs when a user clicks on an ad, is referred to the advertiser's web page, and consummates a purchase before leaving that web page. Alternatively, a conversion may be defined as a user being shown an ad, and making a purchase on the advertiser's web page within a predetermined time (e.g., seven days). Many other definitions of what constitutes a conversion are possible. For example, a conversion also may include signing up to be a member of a web site, filling out an online form, making a purchase upon contacting an advertiser via online creatives and the like.” Looking at historical conversion activities implies that the conversion activities are related to user interactions with a web page.). Sharma does not explicitly disclose wherein the model comprises a page section corresponding to a webpage identifier where a user viewed a content item. Sharma uses machine learning to update the models related to ads, auctions, etc. and the learning may utilize historical data. Ads may be viewed on a website. As seen above, Sharma also acknowledges that there is optimal ad placement on a website. Sharma simply does not explicitly explain that the model itself takes into account where a user viewed a content item (e.g., to address the page section aspects of the model). Fernandez-Ruiz explicitly describes how a model may be trained and optimized based on specific user engagement metrics, including a clickthrough rate and an advertisement conversion rate and the engagement analysis is used in conjunction with a page layout optimizer (Fernandez-Ruiz: ¶ 62). The Examiner submits that it would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s invention to modify Sharma wherein the model comprises a page section corresponding to a webpage identifier where a user viewed a content item in order to allow for continuous updates of the learning model in light of actual user activities so that the model can remain as accurate as possible over time, especially as it pertains to ad placement. Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (US 2008/0275757) in view of Lv et al. (CN-110570232-A, citing the English Translation) in view of Kaufman et al. (US 2024/0370733) in view of Dirac et al. (US 2015/0379430), as applied to claim 1 above, in view of Lo et al. (US 2018/0321980). [Claim 7] Sharma discloses adjusting, during training of the model, predictions generated by the model based upon the model under-predicting (¶ 61 – “ In some instances, these factors may cause the pCVR to over-predict or under-predict a conversion rate. The inaccuracy can lead to a deviation between the revenues generated by the publishers (i.e., eCPC). For example, if the pCVR is over-predicted (e.g., from 0.1% to 0.2%), then a publisher may be credited at a higher cost per click than an actual cost for each displayed ad. As another example, if the pCVR is under-predicted (e.g., from 0.1% to 0.05%), the publisher may be compensated with a lower amount than that would have received under the designated pricing model.”). Sharma does not explicitly disclose that the model under-predicting is due to the model being trained on an additional negative event for each positive event. However, Lo describes problems that can occur when there are more negative errors that positive errors: [0161] At 104, execution time of the plurality of program tasks on one or more computing cores is estimated. Each program feature is mapped to an execution time estimate on a selected computing core. Predicting execution time of the program features involves training and fitting a mathematical model. In some embodiments, a linear model trained with offline training data sets can achieve this purpose. A linear model has the advantage of being simple to train and fast to evaluate at runtime. The most common way to fit a linear model is to use linear least squares regression, aiming to minimize the sum of the absolute errors in the prediction. However, negative and positive errors should be weighed differently for the purpose of meeting response-time requirements. Negative errors (under-prediction) lead to deadline misses that would impact user experience, while positive errors (over-prediction) result in an overly conservative frequency setting which does not save as much energy as possible. To maintain a good user experience, the training process should place greater weight on avoiding under-prediction as opposed to over-prediction. The Examiner submits that it would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s invention to modify Sharma such that the model under-predicting is due to the model being trained on an additional negative event for each positive event so that Sharma can make model corrections accordingly to increase modeling accuracy, which is a goal of Sharma, as evidenced by its use of a correction factor to account for over-prediction or under-prediction (as seen in ¶¶ 61-63 of Sharma). Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (US 2008/0275757) in view of Lv et al. (CN-110570232-A, citing the English Translation) in view of Kaufman et al. (US 2024/0370733) in view of Dirac et al. (US 2015/0379430), as applied to claim 1 above, in view of Chang et al. (US 2009/0254410). [Claim 8] Sharma discloses calculating the target cost per action, wherein the products are part of retargeting impressions (¶ 10 – “An advertiser specifies a target bid (e.g., CPA target bid or other target) for a conversion event associated with an ad. A predicted conversion rate or value is determined (e.g., empirically) for potential impressions of the ad based on conversion data (e.g., historical conversion data) for the ad and impression context data (e.g., current impression context data). The predicted conversion rate and target bid can be used to estimate a click-based bid. Publishers can be compensated based on the estimated click-based bid, while advertisers can be debited using the target bid originally specified.” Considering that an ad may have multiple impressions, this implies that the ads may be retargeted impressions. The ads may be correlated with products (¶ 34). By nature of the association of the products with the ads, a ranking of the ads and conversion rates may be seen as also reflecting a ranking of the advertised products.; ¶ 27 – “In some implementations, when the user 218 clicks an ad served by the ad server 206, the user 218 is directed to a landing page on the web property (e.g., a web site) of the advertiser 214. The user 218 may then perform a conversion event at the website (e.g., make a purchase, register). The conversion event generates conversion data which is sent to the system 200 and stored in a repository (e.g., MySQL® database). In this manner, a conversion history can be accumulated and maintained for each ad or ad group in an advertiser's ad campaign.”). Sharma does not explicitly disclose calculating the target cost per action by applying a factor to an average spend divided by a number of conversions of products of a content provider of the product. Chang explains that “the volume of clicks corresponds to the number of clicks seen in a core market, the conversion rate corresponds to an expected conversion rate for the key words that make up the core market description 200, and the average cost per acquisition (CPA) corresponds to the average amount an advertiser spends divided by a number of sales the advertiser makes. It may be necessary to accurately compute the conversion rate so as to minimize the risk to the sponsored search provider.” (Chang: ¶ 33) The Examiner submits that it would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s invention to modify Sharma to perform the step of calculating the target cost per action by applying a factor to an average spend divided by a number of conversions of products of a content provider of the product is order to incorporate a more accurate computation of cost per action so that advertisers and people selling ad space can plan their bidding and pricing strategies in a more effective and efficient manner. Claims 14-15 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (US 2008/0275757) in view of Lv et al. (CN-110570232-A, citing the English Translation) in view of Kaufman et al. (US 2024/0370733), as applied to claim 12 above, in view of Liu et al. (US 2015/0339704). [Claims 14-15] Sharma does not explicitly disclose: [Claim 14] reducing the target cost per action to reduce a value of the bid and to decrease a pool of eligible users for the product; [Claim 15] increasing the target cost per action to increase the value of the bid and to increase a pool of eligible users for the product. Liu explains: [0091] AMS 106 generates auction scores for a predefined number of advertisements to develop 314 a pool of candidate advertisements for presentation to a publisher. Typically, the candidate advertisements are ranked according to auction score. Accordingly, because auction score is directly proportional to CPA price, the higher a price that advertiser 102 is willing to pay, the higher the auction score for the advertisement is likely to be. The methods described herein enable an advertiser 102 to value their advertisements in relation to their estimated lifetime revenues, relative to the specific consumers to whom the advertisements may be presented. In other words, there is value attributed to advertisements that can increase lifetime revenues from users who engage in sales conversions in response to the selected ads. High cost per action is associated with higher value; however, higher value ads are expected to yield more lifetime revenue and vice-versa. The Examiner submits that it would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s invention to modify Sharma to perform the steps of: [Claim 14] reducing the target cost per action to reduce a value of the bid and to decrease a pool of eligible users for the product; [Claim 15] increasing the target cost per action to increase the value of the bid and to increase a pool of eligible users for the product in order to encourage higher value ad bids for ads that are more likely to increase lifetime revenues and to encourage lower value ad bids for ads that are less likely to yield revenue, thereby enabling advertisers to better strategize when and how to spend their marketing budget. Claims 18-19 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (US 2008/0275757) in view of Kaufman et al. (US 2024/0370733) in view of Dirac et al. (US 2015/0379430), as applied to claim 16 above, in view of Liu et al. (US 2015/0339704). [Claims 18-19] Sharma does not explicitly disclose: [Claim 18] reducing the target cost per action to reduce a value of the bid and to decrease a pool of eligible users for the product; [Claim 19] increasing the target cost per action to increase the value of the bid and to increase a pool of eligible users for the product. Liu explains: [0091] AMS 106 generates auction scores for a predefined number of advertisements to develop 314 a pool of candidate advertisements for presentation to a publisher. Typically, the candidate advertisements are ranked according to auction score. Accordingly, because auction score is directly proportional to CPA price, the higher a price that advertiser 102 is willing to pay, the higher the auction score for the advertisement is likely to be. The methods described herein enable an advertiser 102 to value their advertisements in relation to their estimated lifetime revenues, relative to the specific consumers to whom the advertisements may be presented. In other words, there is value attributed to advertisements that can increase lifetime revenues from users who engage in sales conversions in response to the selected ads. High cost per action is associated with higher value; however, higher value ads are expected to yield more lifetime revenue and vice-versa. The Examiner submits that it would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s invention to modify Sharma to perform the steps of: [Claim 18] reducing the target cost per action to reduce a value of the bid and to decrease a pool of eligible users for the product; [Claim 19] increasing the target cost per action to increase the value of the bid and to increase a pool of eligible users for the product in order to encourage higher value ad bids for ads that are more likely to increase lifetime revenues and to encourage lower value ad bids for ads that are less likely to yield revenue, thereby enabling advertisers to better strategize when and how to spend their marketing budget. Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (US 2008/0275757) in view of Kaufman et al. (US 2024/0370733) in view of Dirac et al. (US 2015/0379430), as applied to claim 16 above, in view of Chang et al. (US 2009/0254410). [Claim 20] Sharma discloses calculating the target cost per action, wherein the products are part of retargeting impressions (¶ 10 – “An advertiser specifies a target bid (e.g., CPA target bid or other target) for a conversion event associated with an ad. A predicted conversion rate or value is determined (e.g., empirically) for potential impressions of the ad based on conversion data (e.g., historical conversion data) for the ad and impression context data (e.g., current impression context data). The predicted conversion rate and target bid can be used to estimate a click-based bid. Publishers can be compensated based on the estimated click-based bid, while advertisers can be debited using the target bid originally specified.” Considering that an ad may have multiple impressions, this implies that the ads may be retargeted impressions. The ads may be correlated with products (¶ 34). By nature of the association of the products with the ads, a ranking of the ads and conversion rates may be seen as also reflecting a ranking of the advertised products.; ¶ 27 – “In some implementations, when the user 218 clicks an ad served by the ad server 206, the user 218 is directed to a landing page on the web property (e.g., a web site) of the advertiser 214. The user 218 may then perform a conversion event at the website (e.g., make a purchase, register). The conversion event generates conversion data which is sent to the system 200 and stored in a repository (e.g., MySQL® database). In this manner, a conversion history can be accumulated and maintained for each ad or ad group in an advertiser's ad campaign.”). Sharma does not explicitly disclose wherein the operations comprise: calculating the target cost per action by applying a factor to an average spend divided by a number of conversions of products of a content provider of the product. Chang explains that “the volume of clicks corresponds to the number of clicks seen in a core market, the conversion rate corresponds to an expected conversion rate for the key words that make up the core market description 200, and the average cost per acquisition (CPA) corresponds to the average amount an advertiser spends divided by a number of sales the advertiser makes. It may be necessary to accurately compute the conversion rate so as to minimize the risk to the sponsored search provider.” (Chang: ¶ 33) The Examiner submits that it would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s invention to modify Sharma wherein the operations comprise: calculating the target cost per action by applying a factor to an average spend divided by a number of conversions of products of a content provider of the product is order to incorporate a more accurate computation of cost per action so that advertisers and people selling ad space can plan their bidding and pricing strategies in a more effective and efficient manner. Conclusion Any inquiry concerning this communication or earlier communications from the examiner should be directed to SUSANNA M DIAZ whose telephone number is (571)272-6733. The examiner can normally be reached M-F, 8 am-4:30 pm. Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Brian Epstein can be reached at (571) 270-5389. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /SUSANNA M. DIAZ/ Primary Examiner Art Unit 3625A