DETAILED ACTION
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 Election/Restrictions
The applicant has elected Group I (claims 1-8 and 17-20) in response to the Requirement for Restriction dated February 25, 2026. Thus, claim 9-16 are withdrawn from further consideration pursuant to 37 CFR 1.142(b) as being drawn to a nonelected invention, there being no allowable generic or linking claim. Election was made without traverse in the reply filed on April 17, 2026.
Response to Amendment
The amendment filed on April 17, 2026 cancelled no claims. No claims were amended, no new claims were added, and claims 9-16 were withdrawn. Therefore, the currently pending claims addressed below are claims 1-8 and 17-20.
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 1-8 and 17-20 are directed to a method and an apparatus which would be classified under one of the listed statutory classifications (i.e., 2019 Revised Patent Subject Matter Eligibility Guidance (hereinafter “PEG”) “PEG” Step 1=Yes).
However, claims 1-8 and 17-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Claim(s) 1 and 17 recite(s) the following abstract idea:
obtaining a first sample set, wherein the first sample set comprises a first training sample, and the first training sample comprises at least one advertisement feature of an advertisement and at least one context feature corresponding to the advertisement;
processing the at least one context feature by using a first model, to obtain a first feature vector; and
processing the at least one advertisement feature by using a second model, to obtain a second feature vector;
processing a processing result of at least one stage in the first model and a processing result of at least one stage in the second model by using a third model, to obtain a first correlation score, wherein the first correlation score represents a degree of matching between the at least one context feature and the at least one advertisement feature;
obtaining, based on the first feature vector, the second feature vector, the first correlation score, and a sample label of the first training sample, a loss corresponding to the first training sample; and
training the first model and the second model based on a loss corresponding to at least one training sample in the first sample set.
The limitations as detailed above, as drafted, falls within the “Certain Method of Organizing Human Activity” grouping of abstract ideas namely commercial or legal interactions because they recite advertising marketing and sales activities or behaviors by training a model for the purpose of improving advertisement placement effect. Accordingly, the claim recites an abstract idea (i.e. “PEG” Revised Step 2A Prong One=Yes).
This judicial exception is not integrated into a practical application because the claim only recites the additional elements of a general-purpose computer comprising one or more processors and a memory storing instructions (e.g., a general-purpose computer with generic computer components) and executing generic neural network model software (e.g., a generic computer component).
The following limitations, if removed from the abstract idea and considered additional elements, merely perform generic computer function of processing, storing, communicating (e.g., transmitting and receiving), and displaying data and, as such, are insignificant extra-solution activities (see MPEP 2016.05(d)(II) and MPEP 2106.05(g)):
obtaining a first sample set, wherein the first sample set comprises a first training sample, and the first training sample comprises at least one advertisement feature of an advertisement and at least one context feature corresponding to the advertisement (receiving data).
The additional technical elements above are recited at a high-level of generality (i.e. as a generic processor performing a generic computer function of processing, communicating and displaying) such that it amounts to no more than mere instructions to apply the exception using a generic computer component. The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional technical elements above do not integrate the abstract idea/judicial exception into a practical application because it does not impose any meaningful limits on practicing the abstract idea. More specifically, the additional elements fail to include (1) improvements to the functioning of a computer or to any other technology or technical field (see MPEP 2106.05(a)), (2) applying or using a judicial exception to effect a particular treatment or prophylaxis for a disease or medical condition (see Vanda memo), (3) applying the judicial exception with, or by use of, a particular machine (see MPEP 2106.05(b)), (4) effecting a transformation or reduction of a particular article to a different state or thing (see MPEP 2106.05(c)), or (5) applying or using 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 (see MPEP 2106.05(e) and Vanda memo).
Rather, the limitations merely add the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (see MPEP 2106.05(f)), or generally link the use of the judicial exception to a particular technological environment or field of use (see MPEP 2106.05(h)).
Thus, the claim is “directed to” an abstract idea (i.e. “PEG” Revised Step 2A Prong Two=Yes)
When considering Step 2B of the Alice/Mayo test, the claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the claims do not amount to significantly more than the abstract idea.
More specifically, as discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using a general-purpose computer comprising one or more processors and a memory storing instructions (e.g., a general-purpose computer with generic computer components and executing generic neural network model software (e.g., a generic computer component) to perform the claimed functions amounts to no more than mere instructions to apply the exception using a generic computer component.
“Generic computer implementation” is insufficient to transform a patent-ineligible abstract idea into a patent-eligible invention (See Affinity Labs, _F.3d_, 120 U.S.P.Q.2d 1201 (Fed. Cir. 2016), citing Alice, 134 S. Ct. at 2352, 2357) and more generally, “simply appending conventional steps specified at a high level of generality” to an abstract idea does not make that idea patentable (See Affinity Labs, _F.3d_, 120 U.S.P.Q.2d 1201 (Fed. Cir. 2016), citing Mayo, 132 S. Ct. at 1300). Moreover, “the use of generic computer elements like a microprocessor or user interface do not alone transform an otherwise abstract idea into patent-eligible subject matter (See FairWarning, 120 U.S.P.Q.2d. 1293, citing DDR Holdings, LLC v. Hotels.com, L.P., 773 F.3d 1245, 1256 (Fed. Cir. 2014)). As such, the additional elements of the claim do not add a meaningful limitation to the abstract idea because they would be generic computer functions in any computer implementation. Thus, taken alone, the additional elements do not amount to significantly more than the above-identified judicial exception (the abstract idea). Looking at the limitations as an ordered combination adds nothing that is not already present when looking at the elements taken individually. There is no indication that the combination of elements improves the functioning of the computer or improves any other technology. Their collective functions merely provide generic computer implementation.
The Examiner notes simply implementing an abstract concept on a computer, without meaningful limitations to that concept, does not transform a patent-ineligible claim into a patent-eligible one (See Accenture, 728 F.3d 1336, 108 U.S.P.Q.2d 1173 (Fed. Cir. 2013), citing Bancorp, 687 F.3d at 1280), limiting the application of an abstract idea to one field of use does not necessarily guard against preempting all uses of the abstract idea (See Accenture, 728 F.3d 1336, 108 U.S.P.Q.2d 1173 (Fed. Cir. 2013), citing Bilski, 130 S. Ct. at 3231), and further the prohibition against patenting an abstract principle “cannot be circumvented by attempting to limit the use of the [principle] to a particular technological environment” (See Accenture, 728 F.3d 1336, 108 U.S.P.Q.2d 1173 (Fed. Cir. 2013), citing Flook, 437 U.S. at 584), and finally merely limiting the field of use of the abstract idea to a particular existing technological environment does not render the claims any less abstract (See Affinity Labs, _F.3d_, 120 U.S.P.Q.2d 1201 (Fed. Cir. 2016), citing Alice, 134 S. Ct. at 2358; Mayo, 132 S. Ct. at 1294; Bilski v. Kappos, 561 U.S. 593, 612 (2010); Content Extraction & Transmission LLC v. Wells Fargo Bank, Nat’l Ass’n, 776 F.3d 1343, 1348 (Fed. Cir. 2014); buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355 (Fed. Cir. 2014).
Applicant herein only requires a general-purpose computer (as evidenced from at least paragraphs 44-47 and 160-167 of the applicant’s specification which disclose that the processor and memory a generic computer components and at least paragraphs 12 and 21 of the applicant’s specification which discloses that the neural network can be one of a plurality of known neural networks and the neural network is not some type of new neural network invented by the applicant); therefore, there does not appear to be any alteration or modification to the generic activities indicated, and they are also therefore recognized as insignificant activity with respect to eligibility. Finally, the following limitations, if removed from the abstract idea and considered additional elements, would be considered insignificant extra solution activity as they are directed to merely receiving, displaying, storing, and/or transmitting data (see MPEP 2016.05(d)(II) and MPEP 2106.05(g)):
obtaining a first sample set, wherein the first sample set comprises a first training sample, and the first training sample comprises at least one advertisement feature of an advertisement and at least one context feature corresponding to the advertisement (receiving data).
Thus, taken individually and in combination, the additional elements do not amount to significantly more than the above-identified judicial exception (the abstract idea) (i.e. “PEG” Step 2B=No).
Dependent claims 2-8 and 18-20 appear to merely further limit the abstract idea by further limiting the processing of the at least one context feature by using a first model, to obtain a first feature vector which is considered part of the abstract idea (Claims 2 and 18); further limiting the processing of the at least one advertisement feature by using a second model, to obtain a second feature vector which is considered part of the abstract idea (Claims 3 and 19); further limiting the obtaining, based on the first feature vector, the second feature vector, the first correlation score, and a sample label of the first training sample, a loss corresponding to the first training sample which is considered part of the abstract idea (Claims 4 and 20); further limiting the obtaining of the first loss based on the first feature vector, the second feature vector, and the sample label of the first training sample which is considered part of the abstract idea (Claim 5); adding an additional step of training a third model which is considered part of the abstract idea (Claim 6); further limiting the third module which is considered part of the abstract idea and adding an additional element of a neural network model which has already been addressed above (Claim 7); and further limiting the first model and the second model which is considered part of the abstract idea (Claim 8), and therefore only further limit the abstract idea (i.e. “PEG” Revised Step 2A Prong One=Yes), does/do not include any new additional elements that are sufficient to amount to significantly more than the judicial exception, and as such are “directed to” said abstract idea (i.e. “PEG” Step 2A Prong Two=Yes); and do not add significantly more than the idea (i.e. “PEG” Step 2B=No).
Thus, based on the detailed analysis above, claims 1-8 and 17-20 are not patent eligible.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 2 and 18 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 2 and 18, depend from claims 1 and 17 respectively, and recite “: The method according to claim 1 “wherein the processing the at least one context feature by using a first model, to obtain a first feature vector comprises: encoding the at least one context feature by using the first model, to obtain N third feature vectors, wherein N>=1”.
First, this limitation changes the scope of the claims from which they depend by allowing the claims “a first feature vector”, which is singular, to be two or more “third feature vectors”, which is plural. Thus, it would appear that the claims change the scope of the claims from which they depend rather than further limiting the scope of the claims. One of ordinary skill in the art would not be able to determine how the applicant’s invention can perform this step without changing the scope of the previously claim “a first feature vector”.
Second, the next limitation in the claims is “processing, when N=1, the N third feature vectors by using the first model, to obtain the first feature vector”. Assuming, that the encoding step in claims 2 and 18 resulted in a single third feature vector being obtained (i.e., N=1), the claims require that this single third feature vector be processed, by the first model, to obtain the first feature vector which changes the scope of claims 1 and 17, rather than further limiting the scope. Claims 1 and 17, from which claims 2 and 18 depend, require the processing of the context feature be done by the first model to obtain the first feature vector. However, claims 2 and 18, indicate that the first feature vector is not obtained by processing, by the first model, the context feature. Instead, it is obtained by processing, by the first model, the third feature vector, wherein the third feature vector was obtained based on the content feature being processed by the first model. One of ordinary skill in the art would not be able to determine how the applicant’s invention can perform this step without changing the scope of the previously claimed “a first feature vector” which was obtained by the first model using the at least one context feature as required by claims 1 and 17.
Finally, claims 2 and 18 recite “or when N>2, concatenating the N third feature vectors by using the first model, to obtain one fourth feature vector”. Once again, this limitation appears to change the scope of claims 1 and 17 from which claims 2 and 18 depend. Claims 1 and 17 require that the invention obtain a first feature vector by processing, using the first model, the at least one context feature. However, the above limitation requires that instead of obtaining the first feature vector as required by claims 1 and 17, the invention obtains two or more third feature vectors by using the first model to process the at least one context feature; concatenating, by the first model, the three or more third feature vectors into a fourth feature vector; and processing the fourth feature vector, using the first model, to obtain the first feature vector. As such, the first feature vector is not obtained by using the first model to process the at least one context feature. Instead, the first feature vector is obtained, using the first model, by processing the concatenated fourth feature vector output by an initial processing, using the first model, of the at least one context feature. One of ordinary skill in the art would not be able to determine how the applicant’s invention can perform this step without changing the scope of the previously claimed “a first feature vector” which was obtained by the first model using the at least one context feature as required by claims 1 and 17.
As such, it is clear that claims 12 and 18 are indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor, regards as the invention.
For the purpose of prosecuting the claims, the examiner is going to interpret claims 1 and 17 as the generating of the first feature vector is obtained, using the first model, based on the at least one content feature to be N=1. Thus, the third feature vector of claims 2 and 18 is the first feature vector. While, claims 2 and 18 would appear to indicate that this would mean that the first feature vector is processed by the first model to output the first feature vector, such processing could not change the first feature vector because it has antecedent basis to the first feature vector which was processed. Given that the first feature vector cannot be changed in any way by processing it through the first model, this additional processing step would appear to serve no practical function and, as such, would not appear to limit the scope of the claims.
Claims 3 and 19 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 3 and 19, depend from claims 1 and 17 respectively, and recite “: The method according to claim 1 “wherein the processing the at least one advertisement feature by using a second model, to obtain a second feature vector comprises: encoding the at least one advertisement feature by using the second model, to obtain M fifth feature vectors, wherein M>=1”.
First, this limitation changes the scope of the claims from which they depend by allowing the claimed “a second feature vector”, which is singular, to be two or more “fifth feature vectors”, which is plural. Thus, it would appear that the claims change the scope of the claims from which they depend rather than further limiting the scope of the claims. One of ordinary skill in the art would not be able to determine how the applicant’s invention can perform this step without changing the scope of the previously claim “a second feature vector”.
Second, the next limitation in the claims is “processing, when M=1, the M third feature vectors by using the second model, to obtain the second feature vector”. Assuming, that the encoding step in claims 2 and 18 resulted in a single fifth feature vector being obtained (i.e., M=1), the claims require that this single fifth feature vector be processed, by the second model, to obtain the second feature vector which changes the scope of claims 1 and 17, rather than further limiting the scope. Claims 1 and 17, from which claims 2 and 18 depend, require the processing of the advertisement feature be done by the second model to obtain the second feature vector. However, claims 2 and 18, indicate that the second feature vector is not obtained by processing, by the second model, the advertisement feature. Instead, it is obtained by processing, by the second model, the fifth feature vector, wherein the fifth feature vector was obtained based on the advertisement feature being processed by the second model. One of ordinary skill in the art would not be able to determine how the applicant’s invention can perform this step without changing the scope of the previously claimed “a second feature vector” which was obtained by the second model using the at least one advertisement feature as required by claims 1 and 17.
Finally, claims 2 and 18 recite “or when M>2, concatenating the M fifth feature vectors by using the second model, to obtain one sixth feature vector”. Once again, this limitation appears to change the scope of claims 1 and 17 from which claims 2 and 18 depend. Claims 1 and 17 require that the invention obtain a second feature vector by processing, using the second model, the at least one advertisement feature. However, the above limitation requires that instead of obtaining the second feature vector as required by claims 1 and 17, the invention obtains two or more fifth feature vectors by using the second model to process the at least one advertisement feature; concatenating, by the second model, the three or more fifth feature vectors into a sixth feature vector; and processing the sixth feature vector, using the second model, to obtain the second feature vector. As such, the second feature vector is not obtained by using the second model to process the at least one advertisement feature. Instead, the second feature vector is obtained, using the second model, by processing the concatenated sixth feature vector output by an initial processing, using the second model, of the at least one advertisement feature. One of ordinary skill in the art would not be able to determine how the applicant’s invention can perform this step without changing the scope of the previously claimed “a second feature vector” which was obtained by the second model using the at least one advertisement feature as required by claims 1 and 17.
As such, it is clear that claims 12 and 18 are indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor, regards as the invention.
For the purpose of prosecuting the claims, the examiner is going to interpret claims 1 and 17 as the generating of the second feature vector is obtained, using the second model, based on the at least one advertisement feature to be M=1. Thus, the fifth feature vector of claims 2 and 18 is the second feature vector. While, claims 2 and 18 would appear to indicate that this would mean that the second feature vector is processed by the second model to output the second feature vector, such processing could not change the second feature vector because it has antecedent basis to the second feature vector which was processed. Given that the second feature vector cannot be changed in any way by processing it through the second model, this additional processing step would appear to serve no practical function and, as such, would not appear to limit the scope of the claims.
Claim Rejections - 35 USC § 102
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.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1-8 and 17-20 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Shi et al. (CN117541324A).
Claims 1 and 17: Shi discloses a model training method and a model training apparatus, comprising:
one or more processors; a memory storing instructions, which when executed by the one or more processors (Paragraphs 17-18, 42-45, and 147-148), cause the apparatus to perform operations:
obtaining a first sample set, wherein the first sample set comprises a first training sample, and the first training sample comprises at least one advertisement feature of an advertisement and at least one context feature corresponding to the advertisement (Paragraphs 13, 74-75, 83, 73, and 50-51: obtaining a first training sample comprising a sample account number information of a sample account number and sample media resource information of a sample media resource the target account number feature vector and the target media resource feature vector of the target account number, wherein the sample information of the sample media includes exposure sample, click sample, non-exposure sample, wherein the exposure sample comprises the sample media with fine ranking score but not published, wherein the sample virtual resource quantity set is the quantity of the virtual resource needed to be transferred by the sample media resource issuing to the sample account number, wherein the sample account information includes demographic data and behavioral data);
processing the at least one context feature by using a first model, to obtain a first feature vector and processing the at least one advertisement feature by using a second model, to obtain a second feature vector (Paragraphs 14, 77, 82, 98-99: the input of the initial account number tower model structure is the sample account number information, the initial account number tower model extracts the characteristic of the sample account number information to obtain the sample account number characteristic vector. The input of the initial media resource tower model structure is the sample media resource, the initial media resource tower model extracts the characteristic of the sample media resource information to obtain the sample media resource characteristic vector); and
processing a processing result of at least one stage in the first model and a processing result of at least one stage in the second model by using a third model, to obtain a first correlation score, wherein the first correlation score represents a degree of matching between the at least one context feature and the at least one advertisement feature (Paragraphs 14, 82, 88, and 98-99: determining the value of the second competition parameter of the ith round and determining the value of the first competition parameter of the ith round and according to the value of the second competition parameter and the actual sample click rate used by the ith round; Paragraphs 93-96, 9-11, 60-69: the manner in which the first and second competition parameter scores are calculated means the first correlation score represents a degree of matching between the at least one context feature and the at least one advertisement feature);
obtaining, based on the first feature vector, the second feature vector, the first correlation score, and a sample label of the first training sample, a loss corresponding to the first training sample; and training the first model and the second model based on a loss corresponding to at least one training sample in the first sample set (Paragraphs 14, 83-86: according to the value of the second competition parameter and the actual sample click rate used by the ith round, determining the first loss value of the ith round, wherein the actual sample click rate used by the ith round is used for representing after the sample media resource used by the ith round is issued to the sample account number used by the ith round, whether the sample account number used by the ith round actually clicks the sample media resource used by the ith round; according to the sample account number characteristic vector determined by the ith round and the sample media resource characteristic vector determined by the ith round, determining the value of the first competition parameter of the ith round, and determining the second loss value of the ith round according to the value of the second competition parameter of the ith round and the value of the first competition parameter of the ith round; determining the target loss value of the ith round according to the first loss value of the ith round and the second loss value of the ith round; when the target loss value of the ith round does not satisfy the target loss condition, adjusting the parameter in at least one of the initial account number tower model structure obtained by the (i-1) th round training and the initial media resource tower model structure obtained by the (i-1) th round training; Paragraph 20: the target prediction neural network model combines the first loss value determined by the sample account number characteristic vector and the sample media resource characteristic vector, and the second loss value determined by the first sample prediction result and the second sample prediction result, the first sample prediction result is the value of the second competition parameter determined according to the sample account number characteristic vector and the sample media resource characteristic vector, the second sample prediction result is according to the sample account number characteristic vector, the sample media resource characteristic vector, and the value of the first competition parameter determined by the corresponding sample virtual resource quantity. That is, in the training process, the first loss value and the second loss value are combined, the training of the model is guided by means of knowledge distillation).
Claims 2 and 18: Shi discloses the method according to claim 1 and the apparatus according to claim 17, wherein the processing the at least one context feature by using a first model, to obtain a first feature vector comprises: encoding the at least one context feature by using the first model, to obtain N third feature vectors, wherein N>=1; and processing, when N=1, the N third feature vectors by using the first model, to obtain the first feature vector; or when N>=2, concatenating the N third feature vectors by using the first model, to obtain one fourth feature vector; and processing the fourth feature vector by using the first model, to obtain the first feature vector. (Paragraphs 14, 77, 82, 98-99: the input of the initial account number tower model structure is the sample account number information, the initial account number tower model extracts the characteristic of the sample account number information to obtain the sample account number characteristic vector. The input of the initial media resource tower model structure is the sample media resource, the initial media resource tower model extracts the characteristic of the sample media resource information to obtain the sample media resource characteristic vector)
Claims 3 and 19: The method according to claim 1 and the apparatus according to claim 17, wherein the processing the at least one advertisement feature by using a second model, to obtain a second feature vector comprises: encoding the at least one advertisement feature by using the second model, to obtain M fifth feature vectors, wherein M>=1; and when M=1, processing the M fifth feature vectors by using the second model, to obtain the second feature vector; or when M>=2, concatenating the M fifth feature vectors by using the second model, to obtain one sixth feature vector; and processing the sixth feature vector by using the second model, to obtain the second feature vector. (Paragraphs 14, 77, 82, 98-99: the input of the initial account number tower model structure is the sample account number information, the initial account number tower model extracts the characteristic of the sample account number information to obtain the sample account number characteristic vector. The input of the initial media resource tower model structure is the sample media resource, the initial media resource tower model extracts the characteristic of the sample media resource information to obtain the sample media resource characteristic vector)
Claims 4 and 20: Shi discloses the method according to claim 1, and the apparatus according to claim 17, wherein the obtaining, based on the first feature vector, the second feature vector, the first correlation score, and a sample label of the first training sample, a loss corresponding to the first training sample comprises: obtaining a first loss based on the first feature vector, the second feature vector, and the sample label of the first training sample; obtaining a second loss based on the first correlation score and the sample label of the first training sample; and obtaining, based on the first loss and the second loss, the loss corresponding to the first training sample. (Paragraphs 14, 83-86: according to the value of the second competition parameter and the actual sample click rate used by the ith round, determining the first loss value of the ith round, wherein the actual sample click rate used by the ith round is used for representing after the sample media resource used by the ith round is issued to the sample account number used by the ith round, whether the sample account number used by the ith round actually clicks the sample media resource used by the ith round; according to the sample account number characteristic vector determined by the ith round and the sample media resource characteristic vector determined by the ith round, determining the value of the first competition parameter of the ith round, and determining the second loss value of the ith round according to the value of the second competition parameter of the ith round and the value of the first competition parameter of the ith round; determining the target loss value of the ith round according to the first loss value of the ith round and the second loss value of the ith round; when the target loss value of the ith round does not satisfy the target loss condition, adjusting the parameter in at least one of the initial account number tower model structure obtained by the (i-1) th round training and the initial media resource tower model structure obtained by the (i-1) th round training; Paragraph 20: the target prediction neural network model combines the first loss value determined by the sample account number characteristic vector and the sample media resource characteristic vector, and the second loss value determined by the first sample prediction result and the second sample prediction result, the first sample prediction result is the value of the second competition parameter determined according to the sample account number characteristic vector and the sample media resource characteristic vector, the second sample prediction result is according to the sample account number characteristic vector, the sample media resource characteristic vector, and the value of the first competition parameter determined by the corresponding sample virtual resource quantity. That is, in the training process, the first loss value and the second loss value are combined, the training of the model is guided by means of knowledge distillation)
Claim 5: Shi discloses the method according to claim 4, wherein the obtaining a first loss based on the first feature vector, the second feature vector, and the sample label of the first training sample comprises:
determining a second correlation score based on the first feature vector and the second feature vector, wherein the second correlation score represents a degree of matching between the at least one context feature and the at least one advertisement feature (Paragraphs 14, 82, 88, and 98-99: determining the value of the second competition parameter of the ith round and determining the value of the first competition parameter of the ith round and according to the value of the second competition parameter and the actual sample click rate used by the ith round; Paragraphs 93-96, 9-11, 60-69: the manner in which the first and second competition parameter scores are calculated means the first correlation score represents a degree of matching between the at least one context feature and the at least one advertisement feature); and
obtaining the first loss based on the second correlation score and the sample label of the first training sample. (Paragraphs 14, 83-86: according to the value of the second competition parameter and the actual sample click rate used by the ith round, determining the first loss value of the ith round, wherein the actual sample click rate used by the ith round is used for representing after the sample media resource used by the ith round is issued to the sample account number used by the ith round, whether the sample account number used by the ith round actually clicks the sample media resource used by the ith round; according to the sample account number characteristic vector determined by the ith round and the sample media resource characteristic vector determined by the ith round, determining the value of the first competition parameter of the ith round, and determining the second loss value of the ith round according to the value of the second competition parameter of the ith round and the value of the first competition parameter of the ith round; determining the target loss value of the ith round according to the first loss value of the ith round and the second loss value of the ith round; when the target loss value of the ith round does not satisfy the target loss condition, adjusting the parameter in at least one of the initial account number tower model structure obtained by the (i-1) th round training and the initial media resource tower model structure obtained by the (i-1) th round training; Paragraph 20: the target prediction neural network model combines the first loss value determined by the sample account number characteristic vector and the sample media resource characteristic vector, and the second loss value determined by the first sample prediction result and the second sample prediction result, the first sample prediction result is the value of the second competition parameter determined according to the sample account number characteristic vector and the sample media resource characteristic vector, the second sample prediction result is according to the sample account number characteristic vector, the sample media resource characteristic vector, and the value of the first competition parameter determined by the corresponding sample virtual resource quantity. That is, in the training process, the first loss value and the second loss value are combined, the training of the model is guided by means of knowledge distillation)
Claim 6: Shi discloses the method according to claim 1, wherein the method further comprises: training the third model based on the loss corresponding to the at least one training sample in the first sample set. (Paragraph 20: in the process of training, the target prediction neural network model, the first loss value and the second loss value are combined, the training of the model is guided by means of knowledge distillation, and the accuracy of the model is improved; the first loss value is determined by the sample account number characteristic vector and the sample media resource characteristic vector; the second loss value determined by the first sample prediction result and the second sample prediction result; the first sample prediction result is the value of the second competition parameter determined according to the sample account number characteristic vector and the sample media resource characteristic vector; the second sample prediction result is according to the sample account number characteristic vector, the sample media resource characteristic vector, and the value of the first competition parameter determined by the corresponding sample virtual resource quantity)
Claim 7: Shi discloses the method according to claim 1, wherein the third model is a neural network model, and is used to assist, in a process of training the first model and the second model, in calculating a loss corresponding to a training sample in the first sample set. (Paragraph 14: the model is a predictive neural network model; adjusting the parameter in at least one of the account number tower model structure obtained by the (i-1)th round training the media resource tower model structure obtained by the (i-1)th round training when the target loss value does not satisfy a target loss condition, wherein the target loss value is determined according to a first loss value of the ith round and a second loss value of the ith round, wherein the first loss value of the ith round and the second loss value of the ith round are associated with first and/or second competition parameters are determined based on the respective vectors output of the first model and/or second model.)
Claim 8: Shi discloses the method according to claim 1, wherein the trained first model is used to perform feature extraction on a context feature related to a page accessed by a user, and the trained second model is used to perform feature extraction on the advertisement. (Paragraphs 73 and 77: the target account tower model structure performs feature extraction on the account information to obtain a feature vector of the target account; the target media resource tower model structure extract features of the media resource information to obtain a feature vector of the media resource; Paragraphs 50-51: the account information includes context features such as the basic portrait feature of the user side, the historical
behavior statistical feature of the user side, the behavior sequence feature of the
user side, and the behavior interest mining feature of the user side, wherein the historical behavioral statistical characteristics include context features of pages access by the user such as the advertisement click number of the user in the last month, the number of clicks of the advertisement detail page in the last month, the video click attention number in the last month, the average exposure advertisement times in the last month; and the behavior sequence features include context features of pages accessed such as recent exposes, clicks, and converts of the advertisement. Paragraph 52: the target media resource may be an advertisement and the features of the target media resource includes a context side feature, an
advertisement side feature. The above context side features include, but are not
limited to, advertising site information (such as advertising site identification,
advertising site material specification, etc.), device information (device operating
system, device networking type), advertising site context information, advertisement identification, creative identification, commodity identification, advertiser identification, advertisement category, creative content keyword, advertisement keyword and so on)
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Wei et al. (CN112446727A) which discloses a neural network model that includes a double tower model, wherein the neural network model determines advertisements to be delivered to a user based on advertisement feature vectors generated by an advertisement tower model and user feature vectors generated by a user tower model, wherein the respective tower models are trained using training sets of historical data, wherein the training sets are split into subsets for the purpose of multi-round training using ith rounds of training, wherein the user context features included page view data, and the advertisements include advertisement features, wherein a correlation score is calculated between the user feature vector and the advertisement feature vector.
Gao et al. (PGPUB: 2015/0363688) which disclose selecting advertisements to be delivered when a user accesses a web page using a deep neural network, wherein advertisements are selected by using a feature vector of the advertisement (e.g., target document) and a feature vector of a web page (e.g., source document), wherein the neural network determines a correlation score (i.e., interestingness score) between the feature vector of the advertisement and the feature vector of the web page, and a loss function is used when training the model to optimize the neural network.
Xu et al., Privileged Features Distillation at Taobao Recommendations, February 26, 2020, https://arxiv.org/pdf/1907.05171, pages 1-9 which discloses fusing an interactive teach model with a dual-tower model during training and optimization which considered a major breakthrough in recommendation systems at the time. In the paper, they describe a unified model architecture where the two-tower student model and an interaction model acting as the privileged or teacher model were trained jointly, where a first loss (classification/cross-entropy loss) is calculated from the output of the user and item tower, and a second loss (a distillation/auxiliary loss) based on the correlation score.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOHN W VAN BRAMER whose telephone number is (571)272-8198. The examiner can normally be reached Monday-Thursday 5:30 am - 4 pm EST.
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, Spar Ilana can be reached at 571-270-7537. 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.
/John Van Bramer/Primary Examiner, Art Unit 3622
Prosecution Insights