US-20260127641-A1 - RECOMMENDATION METHOD AND SYSTEM BASED ON META DATA AUGMENTATION

US20260127641A1US 20260127641 A1US20260127641 A1US 20260127641A1US-20260127641-A1

Abstract

A recommendation method and system based on meta data augmentation relates to the field of personalized recommendation technologies. The method includes: training a cross-domain adaptive encoder-decoder model through user preference data; meta-augmentation ratings of a target domain user-item pair through a trained cross-domain adaptive encoder-decoder model; performing meta-learning training on a recommendation model; and recommending items to a user through a trained recommendation model. In the present disclosure, prior to the meta-learning training of the recommendation model, the cross-domain adaptive encoder-decoder model is trained through the user preference data, and data required for the meta-learning training of the recommendation model is meta-augmentation through the cross-domain adaptive encoder-decoder model, thereby effectively resolving a meta-overfitting problem in the existing meta-learning training of the recommendation model caused by the data sparsity and cold-start problem, and accurately recommending a user's preferred items to the user.

Inventors

Hui Xu
Changyu LI
Yan Zhang
Jie Shao

Assignees

Sichuan Institute of Artificial Intelligence, Yibin, Sichuan, China

Dates

Publication Date: 20260507
Application Date: 20221122
Priority Date: 20210830

Claims (10)

1 - 20 . (canceled)
21 . A computer-implemented method for generating recommendations based on meta data augmentation, the method being performed by a specifically programmed computing device comprising a processor and a memory, the method comprising: S1: receiving, by the computing device, user preference data comprising source domain user-item connection content, target domain user-item connection content, ratings of a source domain user-item pair, and ratings of a target domain user-item pair; and training, by the computing device, a cross-domain adaptive encoder-decoder model using the user preference data, wherein the cross-domain adaptive encoder-decoder model comprises: a first source domain encoder configured to receive ratings of the source domain user-item pair and output a source domain user preference potential representation, a second source domain encoder configured to receive source domain user-item connection content and output a source domain condition term, a first target domain encoder configured to receive ratings of the target domain user-item pair and output a target domain user preference potential representation, a second target domain encoder configured to receive target domain user-item connection content and output a target domain condition term, a source domain decoder configured to receive as input the outputs of the first source domain encoder and the second source domain encoder and reconstruct the ratings of the source domain user-item pair, a target domain decoder configured to receive as input the outputs of the first target domain encoder and the second target domain encoder and reconstruct the ratings of the target domain user-item pair, wherein the encoders and decoders are implemented as neural network modules within the computing device, and the data flow between the encoders and decoders is explicitly defined as above, wherein step S1 comprises the following substeps: S11: encoding the ratings of the source domain user-item pair through the first source domain encoder according to Gaussian distribution N(μ s ,Σ s ) to obtain a source domain user preference potential representation, wherein N( ) is a Gaussian distribution probability density function, μ s is a source domain expectation, and Σ s is a source domain variance, and outputting the source domain user preference potential representation to the source domain decoder; S12: encoding the source domain user-item connection content through the second source domain encoder to obtain a source domain condition term, and outputting the source domain condition term to the source domain decoder; S13: encoding the ratings of the target domain user-item pair through the first target domain encoder according to Gaussian distribution N(μ t ,Σ t ) to obtain a target domain user preference potential representation, wherein μ t is a target domain expectation, and Σ t is a target domain variance, and outputting the target domain user preference potential representation to the target domain decoder; S14: encoding the target domain user-item connection content through the second target domain encoder to obtain a target domain condition term, and outputting the target domain condition term to the target domain decoder; S15: reconstructing, by the source domain decoder, the ratings of the source domain user-item pair using the outputs of the first and second source domain encoders, and reconstructing, by the target domain decoder, the ratings of the target domain user-item pair using the outputs of the first and second target domain encoders; and S16: optimizing the cross-domain adaptive encoder-decoder model by the computing device using a source domain loss function, a target domain loss function, an alternating optimization loss function, and a multi-view information bottleneck constraint object function, each implemented as a neural network loss function executed by the computing device, wherein the multi-view information bottleneck constraint object function is configured to maximize the mutual information between the source domain user preference potential representation and the target domain user preference potential representation while minimizing non-shared information, thereby improving cross-domain adaptation, wherein the alternating optimization loss function is configured to minimize the squared norm between the user preference potential representations and the condition terms for both source and target domains, wherein the above steps are performed by the computing device in a non-generic technical manner to improve recommendation accuracy in sparse data and cold-start scenarios; S2: generating, by the computing device, meta-augmentation ratings for user-item pairs of the target domain by inputting target domain user-item connection content and ratings into the trained cross-domain adaptive encoder-decoder model, and outputting meta-augmentation ratings; S3: constructing, by the computing device, a plurality of meta-learning tasks based on the target domain user-items, target domain ratings, and meta-augmentation ratings, each task comprising a support set and a query set, and performing meta-learning training on a recommendation model using the support set and the query set, wherein the meta-learning training is performed by the computing device to enable rapid adaptation to new users or items; and S4: outputting, by the computing device, a recommendation of one or more items to a user based on the trained recommendation model, wherein the recommendation is generated by executing the trained model on the computing device.
22 . The method of claim 21 , wherein the step of constructing meta-learning tasks (S3) further comprises: S31: constructing, by the computing device, a plurality of task samples for a training task set, each task sample corresponding to a unique target domain user-item connection content and its associated rating, S32: constructing, by the computing device, a plurality of augmented task samples for the training task set, each augmented task sample corresponding to a unique target domain user-item connection content and its associated meta-augmentation rating, S33: sampling, by the computing device, the training task set to obtain a plurality of resampling tasks, and dividing each resampling task into a support sample and a query sample, S34: combining, by the computing device, all support samples to form a support set, and all query samples to form a query set, S35: and performing, by the computing device, inner-loop meta-learning training on the recommendation model using the support set, and S36: performing, by the computing device, outer-loop meta-learning training on the recommendation model using the query set to obtain the trained recommendation model.
23 . A computer system comprising: a domain adaptation subsystem implemented by a processor and memory, the subsystem configured to perform the method of claim 21 , including meta-augmentation of ratings using the cross-domain adaptive encoder-decoder model, and a recommendation subsystem implemented by the processor and memory, the subsystem configured to execute the trained recommendation model to recommend items to users, wherein the system is specifically programmed to perform the steps of claim 21 in a non-generic, technical manner.
24 . The system of claim 23 , wherein the recommendation subsystem comprises a neural network including: a link layer configured to combine user content and item content, at least three hidden layers configured to extract intermediate feature information, an output layer configured to output a user-item recommendation result, wherein the neural network is implemented by the processor and memory of the system.
25 . A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the method of claim 21 .
26 . The computer-readable storage medium of claim 25 , wherein the instructions, when executed, cause the processor to perform the following steps: S1: receiving, by the computing device, user preference data comprising source domain user-item connection content, target domain user-item connection content, ratings of a source domain user-item pair, and ratings of a target domain user-item pair; and training, by the computing device, a cross-domain adaptive encoder-decoder model using the user preference data, wherein the cross-domain adaptive encoder-decoder model comprises: a first source domain encoder configured to receive ratings of the source domain user-item pair and output a source domain user preference potential representation, a second source domain encoder configured to receive source domain user-item connection content and output a source domain condition term, a first target domain encoder configured to receive ratings of the target domain user-item pair and output a target domain user preference potential representation, a second target domain encoder configured to receive target domain user-item connection content and output a target domain condition term, a source domain decoder configured to receive as input the outputs of the first source domain encoder and the second source domain encoder and reconstruct the ratings of the source domain user-item pair, a target domain decoder configured to receive as input the outputs of the first target domain encoder and the second target domain encoder and reconstruct the ratings of the target domain user-item pair, wherein the encoders and decoders are implemented as neural network modules within the computing device, and the data flow between the encoders and decoders is explicitly defined as above, wherein step S1 comprises the following substeps: S11: encoding the ratings of the source domain user-item pair through the first source domain encoder according to Gaussian distribution N(μ s ,Σ s ) to obtain a source domain user preference potential representation, wherein N( ) is a Gaussian distribution probability density function, μ s is a source domain expectation, and Σ s is a source domain variance, and outputting the source domain user preference potential representation to the source domain decoder; S12: encoding the source domain user-item connection content through the second source domain encoder to obtain a source domain condition term, and outputting the source domain condition term to the source domain decoder; S13: encoding the ratings of the target domain user-item pair through the first target domain encoder according to Gaussian distribution N(μ t ,Σ t ) to obtain a target domain user preference potential representation, wherein μ t is a target domain expectation, and Σ t is a target domain variance, and outputting the target domain user preference potential representation to the target domain decoder; S14: encoding the target domain user-item connection content through the second target domain encoder to obtain a target domain condition term, and outputting the target domain condition term to the target domain decoder; S15: reconstructing, by the source domain decoder, the ratings of the source domain user-item pair using the outputs of the first and second source domain encoders; reconstructing, by the target domain decoder, the ratings of the target domain user-item pair using the outputs of the first and second target domain encoders; S16: optimizing the cross-domain adaptive encoder-decoder model by the computing device using a source domain loss function, a target domain loss function, an alternating optimization loss function, and a multi-view information bottleneck constraint object function, each implemented as a neural network loss function executed by the computing device, wherein the multi-view information bottleneck constraint object function is configured to maximize the mutual information between the source domain user preference potential representation and the target domain user preference potential representation while minimizing non-shared information, thereby improving cross-domain adaptation, wherein the alternating optimization loss function is configured to minimize the squared norm between the user preference potential representations and the condition terms for both source and target domains, wherein the above steps are performed by the computing device in a non-generic, technical manner to improve recommendation accuracy in sparse data and cold-start scenarios, S2: generating, by the computing device, meta-augmentation ratings for user-item pairs of the target domain by inputting target domain user-item connection content and ratings into the trained cross-domain adaptive encoder-decoder model, and outputting meta-augmentation ratings; S3; constructing, by the computing device, a plurality of meta-learning tasks based on the target domain user-items, target domain ratings, and meta-augmentation ratings, each task comprising a support set and a query set, and performing meta-learning training on a recommendation model using the support set and the query set, wherein the meta-learning training is performed by the computing device to enable rapid adaptation to new users or items; S4: outputting, by the computing device, a recommendation of one or more items to a user based on the trained recommendation model, wherein the recommendation is generated by executing the trained model on the computing device.
27 . The system of claim 23 , wherein the step of constructing meta-learning tasks (S3) further comprises: S31: constructing, by the computing device, a plurality of task samples for a training task set, each task sample corresponding to a unique target domain user-item connection content and its associated rating, S32: constructing, by the computing device, a plurality of augmented task samples for the training task set, each augmented task sample corresponding to a unique target domain user-item connection content and its associated meta-augmentation rating, S33: sampling, by the computing device, the training task set to obtain a plurality of resampling tasks, and dividing each resampling task into a support sample and a query sample, S34: combining, by the computing device, all support samples to form a support set, and all query samples to form a query set, S35: and performing, by the computing device, inner-loop meta-learning training on the recommendation model using the support set, S36: performing, by the computing device, outer-loop meta-learning training on the recommendation model using the query set to obtain the trained recommendation model.
28 . The device of claim 25 , wherein the instructions, when executed, cause the processor to perform the following steps for constructing meta-learning tasks (S3): S31: constructing, by the processor, a plurality of task samples for a training task set, each task sample corresponding to a unique target domain user-item connection content and its associated rating, S32: constructing, by the processor, a plurality of augmented task samples for the training task set, each augmented task sample corresponding to a unique target domain user-item connection content and its associated meta-augmentation rating, S33: sampling, by the processor, the training task set to obtain a plurality of resampling tasks, and dividing each resampling task into a support sample and a query sample, S34: combining, by the processor, all support samples to form a support set, and all query samples to form a query set, S35: and performing, by the processor, inner-loop meta-learning training on the recommendation model using the support set, S36: performing, by the processor, outer-loop meta-learning training on the recommendation model using the query set to obtain the trained recommendation model.
29 . The computer-readable storage medium of claim 26 , wherein the instructions, when executed, cause the processor to perform the following steps for constructing meta-learning tasks (S3): S31: constructing, by the processor, a plurality of task samples for a training task set, each task sample corresponding to a unique target domain user-item connection content and its associated rating, S32: constructing, by the processor, a plurality of augmented task samples for the training task set, each augmented task sample corresponding to a unique target domain user-item connection content and its associated meta-augmentation rating, S33: sampling, by the processor, the training task set to obtain a plurality of resampling tasks, and dividing each resampling task into a support sample and a query sample, S34: combining, by the processor, all support samples to form a support set, and all query samples to form a query set, S35: performing, by the processor, inner-loop meta-learning training on the recommendation model using the support set, S36: performing, by the processor, outer-loop meta-learning training on the recommendation model using the query set to obtain the trained recommendation model.

Description

CROSS REFERENCE TO THE RELATED APPLICATIONS This application is based upon and claims priority to Chinese Patent Application No. 202111000396.6, filed on Aug. 30, 2021, the entire contents of which are incorporated herein by reference. TECHNICAL FIELD The present disclosure relates to the field of personalized recommendation technologies, and in particular, to a recommendation method and system based on meta data augmentation. BACKGROUND As one of the most critical and effective ways to alleviate information overload, the personalized recommendation technology plays a key role in various applications, such as online e-commerce websites Amazon, Netflix, and Yelp, and online education and news systems. Typically, a recommendation system recommends a personalized list of the most interesting items to a particular user. Existing recommendation systems are mainly based on previous behavior interaction of the user, such as a purchase record, ratings, a click action, a watch record, and others, and therefore are referred to as collaborative filtering (CF) recommendation systems, which have been proved to be very successful. The CF recommendation systems include a user-based CF system that provides interesting, shared items for similar users and an item-based CF system that provides items with similar features for users. However, in practical applications, interaction matrices that represent user behavior interaction are often very sparse because most users have little or even no interaction with items. As a result, useful user preference cannot be effectively learned from limited interaction in the CF recommendation technology, thus resulting in poor performance. Currently, the meta-learning method is mainly adopted to resolve the above problem in the scientific research field. The meta-learning method has a powerful generalization capability and can quickly adapt to new tasks having only a small number of samples. However, in the existing meta-learning recommendation methods, a non-exclusive task is directly constructed from real interaction data and leads to the critical meta-overfitting problem. Therefore, the performance of the existing recommendation method and system is not significantly improved in scenarios of sparse data and cold-start. SUMMARY In view of the foregoing deficiencies in the prior art, the present disclosure provides a recommendation method and system based on meta data augmentation, to improve performance of the existing personalized recommendation technology in scenarios of sparse data and cold-start. To achieve the foregoing objective of the present disclosure, the present disclosure adopts the following technical solutions: According to a first aspect, a recommendation method based on meta data augmentation includes the following steps: S1: training a cross-domain adaptive encoder-decoder model through user preference data;S2: performing meta-augmentation on ratings of a target domain user-item pair through a trained cross-domain adaptive encoder-decoder model to obtain meta-augmentation ratings for user-item pairs of a target domain;S3: constructing different tasks based on target domain user-items, target domain ratings and meta-augmentation ratings in the target domain, each task contains a support set and a query set, and performing meta-learning training on a recommendation model through the support set and the query set; andS4: recommending items to a user through a trained recommendation model. Beneficial effects of the present disclosure are as follows: Prior to the meta-learning training of the recommendation model, the cross-domain adaptive encoder-decoder model is trained through the user preference data, and data required for the meta-learning training of the recommendation model is meta-augmentation through the cross-domain adaptive encoder-decoder model, thereby effectively resolving a meta-overfitting problem in the existing meta-learning training of the recommendation model caused by the sparsity of user and item data and the poor capability to deal with cold-start, and accurately recommending a user's preferred items to the user. Further, the user preference data in step S1 includes source domain user-item connection content, the target domain user-item connection content, ratings of a source domain user-item pair, and the ratings of the target domain user-item pair. Further, the cross-domain adaptive encoder-decoder model includes a first source domain encoder, a second source domain encoder, a first target domain encoder, a second target domain encoder, a source domain decoder, and a target domain decoder. Further, step S1 includes the following substeps: S11: encoding the ratings of the source domain user-item pair through the first source domain encoder according to Gaussian distribution N(μs,Σs) to obtain a source domain user preference potential representation, where N( ) is a Gaussian distribution probability density function, μs is a source domain expectation, an