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 .
DETAILED ACTION
Status of Claims
Claim(s) 1-20 have been examined.
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 claims recite a judicial exception which is not integrated into a practical application and the claims lack an inventive concept.
Step 1 is the first inquiry into eligibility analysis and asks whether the claims are directed to a statutory category. In this instance, the answer must be in the affirmative because they recite a method, medium, and system.
Step 2A prong 1 is the next step in the eligibility analyses and asks whether the claimed invention recites a judicial exception. In this instance, the claims recite the following limitations which comprise the abstract idea:
re-ranking at least some of the plurality of query-item pairs associated with the query that are initially ranked based on the respective query-item pair GBDT output values
This is an abstract idea because it is a mental process which can be performed in one’s mind or using pen and paper.
Step 2A prong 2 is the next step in the eligibility analyses and looks at whether the abstract idea is integrated into a practical application. This requires an additional element or combination of additional elements in the claims to apply, rely on, or use the judicial exception in a manner that imposes a meaningful limit on the judicial exception, such that the claim is more than a drafting effort designed to monopolize the exception.
In this instance, the claims recite the additional elements such as:
obtaining a plurality of query-item pairs associated with a query, wherein the plurality of query-item pairs associated with the query are initially ranked at least partially based on user engagement metrics;
obtaining, for each query-item pair of the plurality of query-item pairs associated with the query, first temporal-behavioral features, second temporal-behavioral features, and context-aware features, wherein the context-aware features include a vertical of the query;
generating, using a Gradient Boosted Decision Tree (GBDT) model, respective query-item pair GBDT output values based on respective inputs for each query-item pair of the plurality of query-item pairs associated with the query, wherein the respective inputs for each query-item pair of the plurality of query-item pairs associated with the query are based on the first temporal-behavioral features, the second temporal-behavioral features, and the context-aware features;
However, these elements do not amount to an improvement in the functioning of a computer or any other technology or technical field, apply the judicial exception with, or by use of, a particular machine, or apply or use the judicial exception in some other meaningful way beyond generally linking the use of the judicial exception to a particular technological environment, such that the claim as a whole is more than a drafting effort designed to monopolize the exception.
In addition, the recitations of these additional limitations are recited at a high level of generality and also do not amount to an improvement in the functioning of a computer or any other technology or technical field, apply the judicial exception with, or by use of, a particular machine, or apply or use the judicial exception in some other meaningful way beyond generally linking the use of the judicial exception to a particular technological environment, such that the claim as a whole is more than a drafting effort designed to monopolize the exception.
The dependent claims also fail to recite elements which amount to an improvement in the functioning of a computer or any other technology or technical field, apply the judicial exception with, or by use of, a particular machine, or apply or use the judicial exception in some other meaningful way beyond generally linking the use of the judicial exception to a particular technological environment, such that the claim as a whole is more than a drafting effort designed to monopolize the exception. For example, claims 2-5, 7-8, and 10 does not amount to an integration according to any one of the considerations above. As for claims 6 and claim 8, these claims, while more specific, still do not amount to an integration according to any one of the considerations above.
Step 2B is the next step in the eligibility analyses and evaluates whether the claims recite additional elements that amount to an inventive concept (i.e., “significantly more”) than the recited judicial exception. According to Office procedure, revised Step 2A overlaps with Step 2B, and thus, many of the considerations need not be re-evaluated in Step 2B because the answer will be the same.
In Step 2A, several additional elements were identified as additional limitations:
obtaining a plurality of query-item pairs associated with a query, wherein the plurality of query-item pairs associated with the query are initially ranked at least partially based on user engagement metrics;
obtaining, for each query-item pair of the plurality of query-item pairs associated with the query, first temporal-behavioral features, second temporal-behavioral features, and context-aware features, wherein the context-aware features include a vertical of the query;
generating, using a Gradient Boosted Decision Tree (GBDT) model, respective query-item pair GBDT output values based on respective inputs for each query-item pair of the plurality of query-item pairs associated with the query, wherein the respective inputs for each query-item pair of the plurality of query-item pairs associated with the query are based on the first temporal-behavioral features, the second temporal-behavioral features, and the context-aware features;
These additional limitations, including the limitations in the dependent claims, do not amount to an inventive concept because they are recited at a high level of generality and also do not amount to an improvement in the functioning of a computer or any other technology or technical field, apply the judicial exception with, or by use of, a particular machine, or apply or use the judicial exception in some other meaningful way beyond generally linking the use of the judicial exception to a particular technological environment, such that the claim as a whole is more than a drafting effort designed to monopolize the exception.
In addition, they were already analyzed under Step 2A and did not amount to a practical application of the abstract idea.
Therefore, the claims lack one or more limitations which amount to an inventive concept in the claims.
For these reasons, the claims are rejected under 35 U.S.C. 101.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
Claim(s) 1-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Sivakumar (US 2022/0398643) in view of Salaka (US 11,269,898) in further view of Reference U (see PTO-892).
Referring to Claim 1, Sivakumar teaches a system comprising: a processor and a non-transitory computer-readable medium storing computing instructions that, when executed on the processor, cause the processor to perform operations comprising:
obtaining a plurality of query-item pairs associated with a query, wherein the plurality of query-item pairs associated with the query are initially ranked at least partially based on user engagement metrics (see Sivakumar ¶¶0055,0041, an item ranking computer device determines an initial set of items responsive to the search query – i.e., query-item pairs, the ranking is determined from user engagement such as clicks and add to cart events);
obtaining, for each query-item pair of the plurality of query-item pairs, behavioral features (see Sivakumar ¶0041, the ranking device determines behavioral features from the engagement data such as clicks and add-to-cart events, associated with the user’s search query and for each query-item pair);
generating, using a gradient boosting model, respective query-item pair GB output values based on respective inputs for each query-item pair of the plurality of query-item pairs associated with the query, wherein the respective inputs for each query-item pair of the plurality of query-item pairs associated with the query are based on the behavioral features (see Sivakumar ¶¶0053,0073, a trained gradient boosting model is applied to features generated for each query-item pair to generate output values used to rank the items);
re-ranking at least some of the plurality of query-item pairs associated with the query that are initially ranked based on the respective query-item pair GB output values (see Sivakumar ¶¶0073-0074, a re-rank model is applied to the re-rank features to generate a re-ranked list of the initial set of items).
Sivakumar teaches obtaining behavioral features (¶0041) and applying a gradient boosting model to rank the query-item pairs (¶¶0053,0073), but does not explicitly teach (i) wherein the behavioral features comprise first temporal-behavioral features and second temporal-behavioral features; (ii) wherein the gradient boosting model is a gradient boosted decision tree; and (iii) obtaining context-aware features, wherein the context-aware features include a vertical of the query.
However, Salaka teaches computing behavioral features over a first predetermined time window and a second predetermined time window (see Salaka Col. 10, behavioral features such as user selections and acquisitions are computed based on a period of time) and that a machine learning model can be a gradient boosted decision tree (see Salaka Col. 10 lines 26-34, Col. 11, Claim 17). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine these references because the results would be predictable. Specifically, the prior art of Sivakumar would still collect behavioral features except that now they would be collected over a time period according to the teachings of Salaka and a gradient boosting model would still be used except that now it would be a gradient boost decision tree model according to the teachings of Salaka. Both of these results are predictable results of the combination.
The combination does not explicitly teach obtaining context-aware features, wherein the context-aware features include a vertical of the query. However, Peng teaches this (see Peng section 3, a shared multi-task BERT model classifies an e-commerce query into its product type/category (i.e., the vertical of the query), and provides the resulting query-category signal to a downstream ranking model). It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Peng with the combination of Sivakumar and Salaka because it would improve ranking relevance and the results of the combination would be predictable. Specifically, providing the product vertical of the query as a context-aware feature to the ranker allows the ranking to be tailored to the category of the query, for which user engagement patterns differ, thereby improving relevance (see Peng section 4, reporting online gains of +2.37% NDCG and +4.82% add-to-cart). In addition, the results would be predictable because Sivakumar already generates features from query and item attributes for the ranking model (¶0056), except that now the feature set would additionally include the context-aware query-vertical feature as taught by Peng. This is a predictable result of the combination.
Referring to Claim 2, the combination teaches the system of claim 1, wherein the operations further comprise: obtaining aggregated historical user engagement data for the query-item pairs associated with the query, wherein the aggregated historical user engagement data for the query-item pairs associated with the query includes, for each query-item pair of the plurality of query-item pairs associated with the query, first user behaviors from a first predetermined time window and second user behaviors from a second predetermined time window the first user behaviors from the first predetermined time window and the second user behaviors from the second predetermined time window include the user engagement metrics (see Sivakumar ¶0041, aggregated per-query engagement including clicks and add-to-cart events; and Salaka Col. 10, user behaviors aggregated “over the last thirty days or a single day”, in other words the first and second user behaviors from several time windows, and the behaviors include user engagement metrics).
Referring to Claim 3, the combination teaches the system of claim 2, wherein obtaining the first temporal-behavioral features and the second temporal-behavioral features further comprises: inputting, to a Bayesian inference framework, the first user behaviors, the first predetermined time window, the second user behaviors, and the second predetermined time window; outputting, by the Bayesian inference framework, the first temporal-behavioral features based on the first user behaviors and the first predetermined time window, and the second temporal-behavioral features based on the second user behaviors and the second predetermined time window (see Salaka Col. 4, the user-interaction data is input to a Bayesian estimation that generates prior and posterior “prediction values” for the behavioral features “in a Bayesian context,” which prediction values are output as the different temporal behavioral features based on the respective user behaviors and predetermined time windows).
Referring to Claim 4, the combination teaches the system of claim 2, wherein obtaining the first temporal-behavioral features and the second temporal-behavioral features further comprises: generating a query string for the query; matching the query string for the query to a stored query string for the query; retrieving the first temporal-behavioral features and the second temporal-behavioral features from a query signal storage repository (see Sivakumar ¶¶0048,0093, the per-item values are stored in memory locations associated with the search query corresponding to “a hash of the search query” and a query item value is later “obtained from a database based on the search query”).
Referring to Claim 5, the combination teaches the system of claim 2, wherein the first predetermined time window is a longer period of time than the second predetermined time window (see Salaka Col. 10, a thirty-day window versus a single-day window).
Referring to Claim 6, the combination teaches the system of claim 1, wherein the operations further comprise: generating the respective inputs by concatenating the first temporal-behavioral features and the second temporal-behavioral features with the context-aware features to generate a combined feature vector for each query-item pair of the plurality of query-items pairs associated with the query; training the GBDT model based on the combined feature vector for each query-item pair of the plurality of query-item pairs associated with the query (see Sivakumar ¶¶0007,0053, features are generated and stored “within a feature vector” and a machine learning model (the gradient boosted decision tree of the combination) is trained with the generated features).
Referring to Claim 7, the combination teaches the system of claim 1, wherein the user engagement metrics include user click metrics, user add-to-cart metrics, and user order metrics (see Sivakumar ¶¶0041,68, item engagements including a number of clicks and a number of add-to-cart events, and user transaction data identifying purchase orders for the items).
Referring to Claim 8, the combination teaches the system of claim 1, wherein the operations further comprise: identifying a triggering condition to perform the re-ranking of at least some of the plurality of query-item pairs associated with the query that are initially ranked based on the query-item pair GBDT output values, wherein the triggering condition includes one or more of a query input by a user, a predetermined time interval elapsing, or a predetermined threshold of new user engagement metrics associated with the query (see Sivakumar ¶0054, ranking is performed responsive to a search query input by a user; and Salaka Cols. 13,15, the ranking values are recomputed on a predetermined interval and the prediction value used is switched after a predetermined time interval elapses).
Referring to Claim 9, the combination teaches the system of claim 1, wherein the operations further comprise: using a multi-task Bidirectional Encoder Representations from Transformers (BERT) model to extract the vertical of the query, wherein the vertical of the query includes one or more of home, food, fashion, or electronics (see Peng section 3, a shared multi-task BERT model performs product type classification and query catalog classification to extract the product category (vertical) of the query; the recited verticals of home, food, fashion, etc., are a subset of such product categories, the selection of which is would have been an obvious design choice).
Referring to Claim 10, the combination teaches the system of claim 6, wherein the GBDT model, as trained, includes one or more vertical nodes and one or more of first temporal-behavioral leaf nodes and second temporal-behavioral leaf nodes of each of the one or more vertical nodes. The gradient boosted decision tree of the combination (see Salaka claim 17) is trained, as described in claims 1 and 6, on a combined feature vector that includes the vertical of the query and the first and second temporal-behavioral features. Accordingly, as inherent of a gradient boosted decision tree, such a tree forms decision (split) nodes on the vertical features (nodes) with terminal leaves reached via splits on the first and second behavioral features (the first and second temporal-behavioral leaf nodes of each vertical node).
Referring to Claim 11, the combination teaches a computer-implemented method comprising:
obtaining, for a plurality of query-item pairs associated with a query, behavioral features, wherein the behavioral features include respective user engagement metrics (see Sivakumar ¶0041, per-query item behavioral features including clicks and add-to-cart engagement metrics);
inputting, to a gradient boosting model, the query and the behavioral features (see Sivakumar ¶¶0052-0053, the generated features are provided to a trained gradient boosting model);
re-ranking at least one query-item pairs of the plurality of query-item pairs associated with the query based on the gradient boosting model output values for the query-item pairs (see Sivakumar ¶¶0073-0074, the initial set of query-item pairs is re-ranked based on the model output values);
Sivakumar does not explicitly teach (i) wherein the behavioral features comprise first temporal-behavioral features for a first predetermined time window and second temporal-behavioral features for a second predetermined time window, each including respective user engagement metrics; (ii) wherein the gradient boosting model is a gradient boosted decision tree. However, Salaka teaches computing behavioral features over a first predetermined time window and second predetermined time window (see Salaka Col. 10) and that the ranking model is a gradient boosted decision tree (see Salaka Col. 10 and Claim 17). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine these references because the results would be predictable. Specifically, the prior art of Sivakumar would still collect behavioral features except that now they would be collected over a time period according to the teachings of Salaka and a gradient boosting model would still be used except that now it would be a gradient boost decision tree model according to the teachings of Salaka. Both of these results are predictable results of the combination.
The combination does not explicitly teach extracting a vertical of the query or inputting the vertical of the query to the model. However, Pen teaches extracting a vertical of a query and providing the vertical of the query to a downstream ranking model (see Peng section 3). It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Peng with the combination of Sivakumar and Salaka because it would improve ranking relevance and the results of the combination would be predictable. Specifically, extracting a product vertical and providing it to the model allows the ranking to be tailored to the category of the query, for which user engagement patterns differ, thereby improving relevance (see Peng section 4, reporting online gains of +2.37% NDCG and +4.82% add-to-cart). In addition, the results would be predictable because Sivakumar already generates features from query and item attributes for the ranking model (¶0056), except that now the feature set would additionally include an extracted query-vertical feature as taught by Peng. This is a predictable result of the combination.
Referring to Claim 12, this claim recites limitations substantially similar to those of claim 4 and is therefore rejected under the same reasons and rationale as set forth in claim 4.
Referring to Claim 13, this claim recites limitations substantially similar to those of claim 2 and is therefore rejected under the same reasons and rationale as set forth in claim 2.
Referring to Claim 14, the combination teaches the computer-implemented method of claim 13, further comprising: retrieving other features for the query-item pairs associated with the query from the query signal storage repository based on the matching the query string for the query to a stored query string for the query, wherein the other features include price, token matching, and popularity (see Sivakumar ¶0069, transaction data identifying “item prices”; ¶0042, comparing “one or more words of the search query to product brands defined within catalog data” is token matching).
Referring to Claim 15, this claim recites limitations substantially similar to those of claim 23 and is therefore rejected under the same reasons and rationale as set forth in claim 3.
Referring to Claim 16, the combination teaches the computer-implemented method of claim 15, wherein the outputting, by the Bayesian inference framework, first temporal-behavioral features and second temporal-behavioral features is performed offline, wherein the first temporal-behavioral features and second temporal-behavioral features are stored offline in the query signal storage repository, and wherein the retrieving the first temporal-behavioral features and the second temporal-behavioral features from a query signal storage repository occurs upon a triggering condition (see Salaka Col. 13, the prior/posterior prediction values are recomputed offline on a periodic basis and stored, retrieved, and applied upon a received query; see Sivakumar ¶0048, the values are stored in the database associated with the search query and retrieved based on the search query).
Referring to Claim 17, this claim recites limitations substantially similar to those of claim 7 and is therefore rejected under the same reasons and rationale as set forth in claim 7.
Referring to Claim 18, this claim recites limitations substantially similar to those of claim 5 and is therefore rejected under the same reasons and rationale as set forth in claim 5.
Referring to Claim 19, the combination teaches the computer-implemented method of claim 11, wherein the plurality of query-item pairs associated with the query are initially ranked by a baseline model, wherein the baseline model is a linear model (see Sivakumar ¶0050, the query-item value that underlies the initial ranking is generated by applying a first weight to the first value, a second weight to the second value, and a third weight to the third value and combining the weighted values, and this weighted-sum is thus linear baseline model, and it produces the ranking of the query-item pairs).
Referring to Claim 20, the combination teaches a non-transitory computer-readable medium storing instructions that upon execution by a processor, cause the processor to perform operations comprising:
obtaining a plurality of query-item pairs associated with a query, wherein the plurality of query-item pairs associated with the query are initially ranked by a baseline model at least partially based on user engagement metrics, token matching, and popularities of each query-item pair of the plurality of query-item pairs (see Sivakumar ¶0050, a weighted sum (linear) baseline model; ¶0041, user engagement metrics such as clicks and add-to-cart events and popularities as a number of engagements normalized against total engagements; ¶0042, token matching by comparing words of the search query to catalog data);
obtaining, for each query-item pair of the plurality of query-item pairs associated with the query, behavioral features from a query signal storage repository (see Sivakumar ¶0048, the per-query-item behavioral features are stored in and retrieved from a database associated with the search query);
generating, using a gradient boosting model, respective query-item pair gradient boosting model output values based on respective inputs for each query-item pair of the plurality of query-item pairs associated with the query, wherein the respective inputs for each query-item pair of the plurality of query-item pairs associated with the query are based the behavioral features (se Sivakumar ¶¶0053,0073, a trained gradient boosting model is applied to features generated for each query-item pair to generate output values used to rank the items);
re-ranking at least some of the plurality of query-item pairs associated with the query that are initially ranked based on the respective query-item pair GBDT output values (see Sivakumar ¶¶0073-0074, a re-rank model is applied to the re-rank features to generate a re-ranked list of the initial set of items).
Sivakumar does not explicitly teach wherein the behavioral features comprise first temporal behavioral features and second temporal behavioral features, or wherein the gradient boosting model is a gradient boosting decision tree. However, Salaka teaches computing behavioral features over a first predetermined time window and a second predetermined time window (see Salaka Col. 10, behavioral features such as user selections and acquisitions are computed based on a period of time) and that a machine learning model can be a gradient boosted decision tree (see Salaka Col. 10 lines 26-34, Col. 11, Claim 17). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine these references because the results would be predictable. Specifically, the prior art of Sivakumar would still collect behavioral features except that now they would be collected over a time period according to the teachings of Salaka and a gradient boosting model would still be used except that now it would be a gradient boost decision tree model according to the teachings of Salaka. Both of these results are predictable results of the combination.
The combination does not explicitly teach obtaining context-aware features, wherein the context-aware features include a vertical of the query. However, Peng teaches this (see Peng section 3, a shared multi-task BERT model classifies an e-commerce query into its product type/category (i.e., the vertical of the query), and provides the resulting query-category signal to a downstream ranking model). It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Peng with the combination of Sivakumar and Salaka because it would improve ranking relevance and the results of the combination would be predictable. Specifically, providing the product vertical of the query as a context-aware feature to the ranker allows the ranking to be tailored to the category of the query, for which user engagement patterns differ, thereby improving relevance (see Peng section 4, reporting online gains of +2.37% NDCG and +4.82% add-to-cart). In addition, the results would be predictable because Sivakumar already generates features from query and item attributes for the ranking model (¶0056), except that now the feature set would additionally include the context-aware query-vertical feature as taught by Peng. This is a predictable result of the combination.
Remarks
Additional prior art relevant to the application but not relied upon includes:
Govindan (US 2021/0233148) teaches smart recommendations for online selections.
Han (US 2014/0280214) teaches multi-phase ranking for content personalization.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MATTHEW E ZIMMERMAN whose telephone number is (571)270-5278. The examiner can normally be reached 8-4pm M-T, 8-12pm W.
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, Marissa Thein can be reached at (571)272-6764. 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.
/MATTHEW E ZIMMERMAN/Primary Examiner, Art Unit 3688
Prosecution Insights