Bilateral personalized federal recommendation method and device, equipment, medium and product

Abstract

The invention discloses a bilateral personalized federal recommendation method and device, equipment, a medium and a product, and relates to the technical field of information recommendation, and the method comprises the steps: determining target recommendation equipment based on an information recommendation instruction; wherein the target recommendation equipment is a user terminal for recommending the article tendency information; the target recommendation device is controlled to send an information calling instruction to the shared server so as to call shared article embedded information stored in the shared server; and determining a first recommendation result of the target recommendation device based on the shared article embedded information and local article embedded information of the target recommendation device, and controlling an interactive interface of the target recommendation device to display the first recommendation result, so that a user can view the first recommendation result. According to the method, the user privacy information and the project public information are classified, processed and stored, the user information is only visible to the user client, the public information is visible to all the clients, and the safety of user data can be ensured as much as possible while the personalized effect of information recommendation is ensured.

Description

Technical Field

[0001] The present invention relates to the field of information recommendation technology, and in particular to a bilateral personalized federated recommendation method, apparatus, device, medium and product. Background Art

[0002] With the development of technology and the increasing demands of daily life, more and more recommendation systems are being widely used. For example, recommendation systems for music, movies, information, and items are being used. These systems can determine the degree of compatibility between various types of recommended information and users based on their usage habits, and recommend highly recommended information to users. Information recommendation primarily uses a centralized data processing approach, which requires uploading user data to a central server for processing. This approach carries a high risk of data leakage, low data security, cumbersome data processing, and limited information recommendation efficiency.

[0003] Therefore, designing an efficient, convenient and secure information recommendation method to recommend information to users in a targeted manner while ensuring the security of user data is an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides a bilateral personalized federated recommendation method, apparatus, device, medium and product, which classifies, processes and stores user privacy information and project public information. User information is visible only to the user client, while public information is visible to all clients. This ensures the personalized effect of project recommendations while ensuring the security of user data as much as possible.

[0005] According to one aspect of the present invention, a bilateral personalized federated recommendation method is provided, the method comprising:

[0006] Based on the information recommendation instruction, determining a target recommendation device; wherein the target recommendation device is a user terminal to which the item tendency information is recommended;

[0007] Controlling the target recommendation device to send an information retrieval instruction to the sharing server to retrieve the shared item embedded information stored on the sharing server;

[0008] Based on the shared item embedding information and the local item embedding information of the target recommendation device, a first recommendation result of the target recommendation device is determined, and the interactive interface of the target recommendation device is controlled to display the first recommendation result so that the user can view the first recommendation result.

[0009] According to another aspect of the present invention, a bilateral personalized federated recommendation apparatus is provided. The bilateral personalized federated recommendation apparatus is configured to implement the bilateral personalized federated recommendation method according to any embodiment of the present invention. The apparatus includes:

[0010] A device determination module is used to determine a target recommendation device based on the information recommendation instruction; wherein the target recommendation device is a user terminal for which item tendency information is recommended;

[0011] an information determination module, configured to control the target recommendation device to send an information retrieval instruction to the sharing server to retrieve the shared item embedded information stored on the sharing server;

[0012] The information recommendation module is used to determine the first recommendation result of the target recommendation device based on the shared item embedding information and the local item embedding information of the target recommendation device, and control the interactive interface of the target recommendation device to display the first recommendation result so that the user can view the first recommendation result.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] at least one processor; and a memory communicatively coupled to the at least one processor;

[0015] The memory stores a computer program that can be executed by at least one processor, and the computer program is executed by at least one processor so that the at least one processor can execute the bilateral personalized federated recommendation method in any embodiment of the present invention.

[0016] According to another aspect of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores computer instructions, which are used to enable a processor to implement the bilateral personalized federated recommendation method in any embodiment of the present invention when executed.

[0017] According to another aspect of the present invention, a computer program product is provided. The computer program product includes a computer program. When the computer program is executed by a processor, it implements the bilateral personalized federated recommendation method of any embodiment of the present invention.

[0018] The bilateral personalized federated recommendation method of the present invention includes: determining a target recommendation device based on an information recommendation instruction; wherein the target recommendation device is a user terminal to which item preference information is recommended; controlling the target recommendation device to send an information retrieval instruction to a shared server to retrieve shared item embedding information stored on the shared server; determining a first recommendation result for the target recommendation device based on the shared item embedding information and the local item embedding information of the target recommendation device, and controlling the interactive interface of the target recommendation device to display the first recommendation result so that the user can view the first recommendation result. The technical solution of the present invention, by classifying and storing user private information and project public information, makes user information visible only to the user client and public information visible to all clients, thereby ensuring the personalized effect of project recommendations while ensuring the security of user data to the greatest extent possible.

[0019] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 This is a flowchart of a bilateral personalized federated recommendation method provided by the present invention;

[0022] Figure 2 1 is a flow chart of another bilateral personalized federated recommendation method provided by the present invention;

[0023] Figure 3 This is a schematic structural diagram of a bilateral personalized federated recommendation device provided by the present invention;

[0024] Figure 4 It is a structural schematic diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first", "second", "initial", "intermediate", "target", "candidate", "alternative", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] Figure 1 This is a flow chart of a bilateral personalized federated recommendation method provided by the present invention. This embodiment is applicable to situations such as recommending information and things of interest to users with high security and high accuracy. The method can be executed by the bilateral personalized federated recommendation device provided by the present invention. The device can be implemented in the form of hardware and / or software. In a specific embodiment, the device can be integrated into an electronic device. The following embodiments will be described using the device integrated into an electronic device as an example. Figure 1 , the method specifically comprises the following steps:

[0028] S101: Determine a target recommended device based on an information recommendation instruction.

[0029] Among them, the information recommendation instruction can be understood as information sent by the user with recommendation needs, which is used to locate the type of recommended information, the device for processing and displaying recommended information, and the type of recommended information includes but is not limited to music recommendations, movie recommendations, information recommendations, item recommendations and interest expansion recommendations. The target recommendation device can be understood as a device that processes and displays recommended information, such as mobile phones, computers, tablets and other devices. It is worth noting that there is a one-to-one correspondence between the device or the user identifier of the device and the user who needs to view the recommended information. When the device is a private device and there is no mixed use or borrowing, the device and the user are directly corresponding, and any instruction triggered by the device can be regarded as the demand of the user corresponding to the device. When the device is a shared device or a device with mixed use and borrowing, there is a corresponding relationship between the user and the account logged in to the device. Any instruction triggered by the device needs to be regarded as the demand of the user corresponding to the account logged in to the device. The purpose of this setting is to recommend personalized information to users in a targeted manner.

[0030] Specifically, the user needs to use the device to trigger the information recommendation instruction. Therefore, the information recommendation instruction will carry the attribute information of the device and the recommendation needs of the user corresponding to the device. The attribute information of the device can be unique information such as the device address, device identification, device serial number, etc. The user's recommendation needs can be music recommendations, movie recommendations, information recommendations, item recommendations, interest expansion recommendations and other information, which is convenient for quickly locating the information recommendation device that processes data and the recommended information range of the information recommendation device, and starting the information recommendation device in a targeted manner, narrowing the data processing range, reducing the data processing volume, and improving the efficiency of information recommendation.

[0031] For example, assuming that the user triggers a movie recommendation instruction through the interactive interface of device A, device A is determined as the target recommendation device, and device A performs the movie recommendation work, and finally the movie recommendation information is displayed on the interactive interface of device A so that the user can understand the movie recommendation situation and watch the movie.

[0032] S102: Control the target recommendation device to send an information retrieval instruction to the sharing server to retrieve the shared item embedded information stored on the sharing server.

[0033] Among them, the shared server can be understood as a cloud server, which can interact with multiple clients, receive public information (item embedding) sent by each client, and send public information to clients with information acquisition needs. The cloud server will update the shared information based on the public information sent by each client, so that all clients and shared servers can download and use unified, high-real-time and dense item embedding. The information retrieval instruction can be understood as the shared information retrieval instruction of the target recommendation device, which can carry the attribute information and recommendation requirements of the target recommendation device. The advantage of this setting is that it can not only narrow the scope of shared information and reduce transmission pressure, but also transmit shared information in a targeted manner to improve information security and transmission efficiency. Shared item embedding information can be understood as the latest public recommendation information stored on the shared server that complies with the information retrieval instruction.

[0034] Specifically, shared item embedding information is used to represent the item attributes of the item to be recommended. Taking movie recommendation as an example, the item to be recommended is the movie. The item attributes of the item to be recommended can be understood as the type information of the movie and the characteristic degree of each type of movie. The type information of the movie includes but is not limited to comedy, thriller, ethics and suspense. The characteristic degree of the movie can be one-dimensional or multi-dimensional. For example, the characteristic degree of movie A is thriller 0.6, and the characteristic degree of movie B is thriller 0.8, suspense 0.3 and ethics 0.2.

[0035] For example, assuming that device A sends a retrieval instruction for movie recommendation information to the shared server, the shared item embedded information retrieved by device A is the latest movie information stored on the shared server, including but not limited to the movie name, movie type, movie poster, movie introduction, movie feature level and movie feature access method, etc.

[0036] S103: Determine a first recommendation result of the target recommendation device based on the shared item embedding information and the local item embedding information of the target recommendation device, and control the interactive interface of the target recommendation device to display the first recommendation result so that the user can view the first recommendation result.

[0037] Among them, local item embedding information is used to represent the user's item preference information, for example, the user's favorite movie type, movie length, movie actors, etc. The first recommendation result can be understood as one or more movies that are more closely matched with the user's viewing habits obtained by combining the shared item embedding information and the user's item preference information (i.e., local item embedding information). That is, based on the user's viewing tendencies, the movie consultations obtained from the shared server are selected and sorted, and movies that the user may be more interested in are displayed, so that the user can quickly obtain and watch movies that are more in line with his or her preferences.

[0038] This invention is a bilaterally personalized federated learning (FL) recommendation method that fully considers the characteristics of globally shared information and the personalized preferences of different users for items. It divides item embeddings into globally shared and locally personalized parts. The globally shared embeddings are uploaded to the cloud server for aggregation and update to learn global item trends and characteristics. The user embeddings (user privacy information) of each client are retained locally for training and update to learn different users' personalized preferences for the same item. When recommending information to users, these two types of information are combined to achieve bilaterally linked personalized recommendations. Each client and server shares a unified and dense item embedding (public information). Private data is stored only on the client and not transmitted externally. This ensures the security of user data while ensuring the effectiveness of item recommendations.

[0039] For example, users can watch movies and click, collect, comment or rate them. These operations reflect the user's preferences in the field of movies to a certain extent. For example, users collect movies of a certain type or certain themes, make positive comments on movies of a certain type or certain themes, and give high scores to movies of a certain type or certain themes. This can be understood as the user's love for watching movies of a certain type or certain themes. The client can recommend movies that suit the interests of specific users based on their preferences, thereby improving ratings and user experience. In order to protect the security of user data, the present invention uses a federated framework, which does not require the disclosure of user preference information to a third party (i.e., a cloud server). It can effectively model movie attributes and recommend movies to users by combining feedback from all users without leaving the local device, while keeping user privacy data within the local device.

[0040] Optionally, before determining the first recommendation result of the target recommendation device based on the shared item embedding information and the local item embedding information of the target recommendation device, the method further includes: processing the shared item embedding information and the local item embedding information using a regularization method so that the shared item embedding information and the local item embedding information are different and complementary; and inducing sparsity of the shared item embedding information to reduce the communication content of the shared item embedding information.

[0041] The present invention adopts a dual regularization strategy. On the one hand, it increases the diversity of shared item embedding information and local item embedding information, making the two types of information complementary. On this basis, to ensure the difference between global and local learning content, a regularization method is also used to maximize the bi-norm between global item embeddings (shared item embedding information) and local item embeddings (local item embedding information). This forces the local item view to learn user-specific information, while the global item view retains universal information. This makes the shared item embedding information and local item embedding information different and complementary, which is more conducive to determining accurate recommendation information. On the other hand, only useful recommendation information is learned, and useless information is adaptively discarded or reduced in weight. The sparsity of the global item embedding is used as a regularization target and constrained by the norm. This induces the sparsity of the shared item embedding information to reduce the communication content of the shared item embedding information. That is, by sparsifying the global embedding, only useful embedding information is transmitted as much as possible, reducing the amount of data exchanged between the server and user devices, thereby reducing the model's communication overhead, improving information transmission efficiency and system response speed, reducing server load, and adapting to multi-user environments.

[0042] Assume that in the process of movie recommendation, there are n users and m movies. Click operation is used as the evaluation basis of user preference (when users browse the movie introduction page, some tags of the movie may attract users, such as funny, suspense, etc., and users will click on the movie to view it in detail). No click is recorded as 0 and click is recorded as 1. n ] T ∈{0,1} n×m Represents the rating data of n users on m movies, r i ∈{0,1} m Represents the rating data of the i-th user. Local model design and training can be performed on the user's mobile device. The model input layer is generally a binary encoding vector converted from the user and movie ID. and For example, for user with user_ID=0 and movie with movie_ID=0, the user binary encoding vector is Movie binary encoding vector Above the input layer is the embedding layer, which converts the sparse representation and Projected into dense vector space, we get user embedding p u and movie embed q i , each dimension of the embedding vector represents some preference information of the user or some attribute information of the movie.

[0043] In order to ensure that local items embed information D (u)The shared item embedding information C is different from the item information learned on user u, and regularization is added to force the differentiation between them, specifically Where n is the number of clients, D (u) It is used to learn user-specific information, C retains universal information that is useful to all users, and F represents the norm. Furthermore, there may be redundant information in C. The present invention also uses the L1 norm to induce the sparsity of C, thereby eliminating unnecessary information and reducing the communication cost between the shared server and the client. The induction method is specifically as follows: The objective function followed by local model training is r ui represents the true rating of user u i, Represents the i-th predicted rating of user u. The function aims to minimize the cross entropy loss between the true rating and the predicted rating, maximize the difference between personalized and shared information in the movie attribute embedding, and minimize the communication content. Minimizing the cross entropy loss between the true rating and the predicted rating can improve the model prediction accuracy, maximizing the difference between personalized and shared information in the movie attribute embedding can ensure information complementarity, and minimizing the communication content can improve communication efficiency.

[0044] Some parameters in the objective function are set and adjusted during the model training and optimization process. Specifically, the user embedding p u and movie embed q i After that, the embedding vector is sent to a multilayer perceptron (MLP) to obtain the prediction score, and a single-layer feedforward neural network is used to learn the interaction between the user and the movie’s potential features. z is the vector obtained by concatenating the user embedding and the movie embedding, and is obtained by a single-layer neural network Φ(z)=a(W T z+b), get an output, and then use the activation function to map the output to a prediction score between 0 and 1 w, b and a represent the weight matrix, bias vector and activation function of the perceptron respectively. The predicted rating is based on p, which represents the user preference feature. u And qi representing the attribute characteristics of the movie is obtained, q i =(D (u) +C) i , D (u) is the local item embedding information, and C is the shared item embedding information.

[0045] Specifically, based on the shared item embedding information and the local item embedding information of the target recommendation device, the first recommendation result of the target recommendation device is determined, including: determining the initial item to be recommended and the first weight value of the initial item to be recommended based on the shared item embedding information; adjusting the first weight value of the initial item to be recommended using the local item embedding information to obtain the second weight value of the initial item to be recommended; using the second weight value and preset filtering conditions to filter the initial item to be recommended to obtain the first recommendation result, where the preset filtering conditions are the preset recommendation threshold and / or the preset weight threshold.

[0046] The initial items to be recommended can be understood as movie resources in a shared server, which can be all or part of the movie resources, and the present invention is not limited to this. Generally, there are multiple items in the initial list of items to be recommended. The first weight value of the initial items to be recommended can be understood as the comprehensive recommendation level score of each type of movie determined based on public information, and the second weight value can be understood as the recommendation level score of each type of movie obtained by adjusting the first weight value based on the user's personalized information.

[0047] Assume that for all users, the film "XX Space" has a thriller score of 0.6 and a suspense score of 0.8. This is the first weight value determined by the globally shared embedding C across all devices. For user A, however, based on the local data on the client, the film's suspense score is only 0.2. This is the second weight value determined by the local embedding D.

[0048] The user side needs to combine the shared embedding C with the user's local data (viewing history, ratings, etc.) for personalized training. For example, if user A likes science fiction movies and user B prefers action movies, during training, features related to science fiction movies will be enhanced for user A's device, while features related to action movies will be added for user B's device. This allows the model to capture each user's unique preferences, rather than relying solely on global trends.

[0049] In addition, local item embedding information can also be used to adjust the weight coefficient of the first weight value. For example, for all users, the performance of the movie "XX Space" in terms of horror features is 0.6, and the performance in terms of suspense features is 0.8. User B needs to adjust the horror features and suspense features to 0.5 times the original based on the general features. For user B, the performance of the movie "XX Space" in terms of horror features is 0.3, and the performance in terms of suspense features is 0.4.

[0050] Using the second weight value and the preset filtering conditions, the initial items to be recommended are filtered, and the first recommendation result can be understood as arranging the items to be recommended according to their feature information, and selecting items with a recommendation threshold higher than the preset weight threshold or a preset recommendation threshold number that are ranked high for display by users for viewing.

[0051] On the one hand, before determining the target recommendation device based on the information recommendation instruction, the present invention also includes: determining the candidate item embedding information, the training device and the recommendation information prediction model stored on the training device, wherein the candidate item embedding information is the general item embedding information stored on the shared server or the latest historical shared item embedding information; controlling the training device to retrieve the candidate item embedding information, optimizing the recommendation information prediction model using the candidate item embedding information and the sample item embedding information stored on the training device, and determining the shared item embedding information and the local item embedding information of the training device based on the model parameters of the trained recommendation information prediction model when the recommendation information prediction model meets the preset training conditions.

[0052] The recommendation information prediction model can be understood as an algorithm for parsing recommendation information, the candidate item embedding information can be understood as a sample for training the recommendation information prediction model, and the training device can be understood as the user end. The present invention will perform model training on the user end so that the successfully trained model can perform the information recommendation task on the user end. The preset training conditions can be understood as the objective function of model training, which is used to make the model converge. It is worth noting that each model training work of the present invention randomly selects a part of users from all user devices to participate in the training. The purpose of this setting is to ensure that the system covers a sufficient number of users and improve the diversity of the model. Optimization and model training can update model parameters in real time, improve the accuracy of public information and personalized information, and thus improve the quality of information recommendation.

[0053] On the other hand, after controlling the interactive interface of the target recommendation device to display the first recommendation result, the present invention also includes: determining the item to be recommended and the predicted evaluation information of the item to be recommended based on the first recommendation result; after receiving the user evaluation information of the user on the recommended item, using the user evaluation information and the predicted evaluation information to optimize the model parameters of the recommendation information prediction model to adjust the shared item embedding information and the local item embedding information of the target recommendation device.

[0054] Determining the recommended item and the predicted evaluation information for the recommended item based on the first recommendation result can be understood as analyzing the first recommendation result and predicting the user's rating of each of the recommended items; and the user evaluation information can be understood as the user's rating of each of the recommended items after learning about the recommended item. After receiving the user's actual evaluation, the present invention also optimizes the recommendation information prediction model based on the difference between the actual evaluation and the predicted evaluation. For example, the prediction weight of feature types where the actual evaluation is lower than the predicted evaluation is reduced, and the prediction weight of feature types where the actual evaluation is higher than the predicted evaluation is increased. This makes the recommendation information prediction model more consistent with the user's usage habits, better recommends information to the user, provides personalized recommendation services, enhances the user experience, and thereby improves user retention rate and usage time.

[0055] Specifically, the user evaluation information and the predicted evaluation information are used to optimize the model parameters of the recommendation information prediction model, including: determining the error items, the type information of the error items and the error data based on the difference information between the user evaluation information and the predicted evaluation information; determining the model parameters to be adjusted of the recommendation information prediction model based on the type information, and adjusting the model parameters to be adjusted using the error data, so that the first matching degree between the recommendation information prediction model and the error items after the model parameters are adjusted is higher than the second matching degree between the recommendation information prediction model and the error items before the model parameters are adjusted.

[0056] Error items can be understood as items with high differences between user ratings and predicted ratings, the type information of error items can be understood as characteristic information of error items, and the error data of error items can be understood as specific evaluation deviation values. The type information and error data of error items can be combined to determine the optimization content and degree of optimization of the recommendation information prediction model. For example, the model parameters to be adjusted that need adjustment can be located based on the type information of the error items, and the adjustment direction and scale of the model parameters to be adjusted can be determined based on the error data of the error items.

[0057] Exemplarily, based on the difference information between user evaluation information and predicted evaluation information, determining the error items, type information of the error items, and error data can be understood as determining movies with a score difference higher than a preset score as model optimization items (i.e., error items), parsing the type information and error data of the error items, and updating the model based on the parsed information, adaptively reducing or increasing the parameter values of the associated parameter items. For example, if the predicted scores of the horror features of the three movies ABC are all higher than the actual scores, then the common information of the three movies is parsed to obtain the model parameters corresponding to the common information, and its weight is adaptively reduced so that when facing the same type of movies in the future, the weight value of the horror feature is reduced, and a movie recommendation result that is more targeted to the user is predicted.

[0058] The technical solution of the above embodiment, through classification, processing and storage of user privacy information and project public information, makes user information visible only to the user client and public information visible to all clients, thereby ensuring the personalized effect of project recommendation while ensuring the security of user data as much as possible.

[0059] Figure 2 This is a flow chart of another bilateral personalized federated recommendation method provided by the present invention. This embodiment provides a preferred information recommendation method based on the above embodiment. Specifically, Figure 2 As shown, the method includes:

[0060] S201: Determine a target recommended device based on an information recommendation instruction.

[0061] The target recommendation device is the user terminal that recommends item tendency information.

[0062] S202: Control the target recommendation device to send an information retrieval instruction to the sharing server to retrieve the shared item embedded information stored on the sharing server.

[0063] S203: Process the shared item embedding information and the local item embedding information using a regularization method, so that the shared item embedding information and the local item embedding information are different and complementary.

[0064] S204: Induce the sparsity of the embedded information of the shared item to reduce the communication content of the embedded information of the shared item.

[0065] S205: Determine the initial items to be recommended and the first weight values of the initial items to be recommended based on the shared item embedding information.

[0066] S206: Adjust the first weight value of the initial item to be recommended by using the local item embedding information to obtain a second weight value of the initial item to be recommended.

[0067] S207: Filter the initial items to be recommended using the second weight value and the preset filtering condition to obtain a first recommendation result.

[0068] Among them, the preset screening condition is a preset recommendation threshold and / or a preset weight threshold.

[0069] S208: Determine the item to be recommended and the predicted evaluation information of the item to be recommended based on the first recommendation result.

[0070] S209: After receiving the user evaluation information of the recommended item, optimize the model parameters of the recommendation information prediction model using the user evaluation information and the predicted evaluation information to adjust the shared item embedding information and the local item embedding information of the target recommendation device.

[0071] The purpose of this setting is to improve the accuracy of information prediction so that the item information recommended by the user terminal is more in line with the user's needs.

[0072] Figure 3 This is a schematic diagram of the structure of a bilateral personalized federated recommendation device provided by the present invention. Figure 3 As shown, the apparatus includes: a device determination module 301 , an information determination module 302 and an information recommendation module 303 .

[0073] The device determination module 301 is used to determine a target recommendation device based on the information recommendation instruction; wherein the target recommendation device is a user terminal to which item tendency information is recommended.

[0074] The information determination module 302 is used to control the target recommendation device to send an information retrieval instruction to the sharing server to retrieve the shared item embedded information stored on the sharing server.

[0075] The information recommendation module 303 is used to determine the first recommendation result of the target recommendation device based on the shared item embedding information and the local item embedding information of the target recommendation device, and control the interactive interface of the target recommendation device to display the first recommendation result so that the user can view the first recommendation result.

[0076] Optionally, the shared item embedding information is used to represent the item attributes of the item to be recommended, and the local item embedding information is used to represent the user's item preference information.

[0077] Optionally, the information recommendation module 303 is further configured to process the shared item embedding information and the local item embedding information of the target recommendation device using a regularization method before determining the first recommendation result of the target recommendation device based on the shared item embedding information and the local item embedding information of the target recommendation device, so that the shared item embedding information and the local item embedding information are different and complementary; and induce the sparsity of the shared item embedding information to reduce the communication content of the shared item embedding information.

[0078] Optionally, the information recommendation module 303 is specifically used to determine the initial items to be recommended and the first weight value of the initial items to be recommended based on the shared item embedding information; adjust the first weight value of the initial items to be recommended using the local item embedding information to obtain the second weight value of the initial items to be recommended; use the second weight value and preset filtering conditions to filter the initial items to be recommended to obtain a first recommendation result, wherein the preset filtering conditions are a preset recommendation threshold and / or a preset weight threshold.

[0079] Optionally, the bilateral personalized federated recommendation device also includes a model training module, which is used to determine the candidate item embedding information, the training device and the recommendation information prediction model stored on the training device before determining the target recommendation device based on the information recommendation instruction, wherein the candidate item embedding information is the general item embedding information stored on the shared server or the latest historical shared item embedding information; control the training device to retrieve the candidate item embedding information, use the candidate item embedding information and the sample item embedding information stored on the training device to optimize the recommendation information prediction model, and when the recommendation information prediction model meets the preset training conditions, determine the shared item embedding information and the local item embedding information of the training device based on the model parameters of the trained recommendation information prediction model.

[0080] Optionally, the model training module is also used to determine the item to be recommended and the predicted evaluation information of the item to be recommended based on the first recommendation result after the interactive interface of the target recommendation device is controlled to display the first recommendation result; after receiving the user evaluation information of the recommended item, the user evaluation information and the predicted evaluation information are used to optimize the model parameters of the recommendation information prediction model to adjust the shared item embedding information and the local item embedding information of the target recommendation device.

[0081] Optionally, a model training module is specifically used to determine error items, type information of error items and error data based on the difference information between user evaluation information and predicted evaluation information; based on the type information, determine the model parameters to be adjusted of the recommendation information prediction model, and use the error data to adjust the model parameters to be adjusted, wherein the first matching degree between the recommendation information prediction model and the error items after adjusting the model parameters is higher than the second matching degree between the recommendation information prediction model and the error items before adjusting the model parameters.

[0082] The bilateral personalized federated recommendation device provided in the above embodiments can execute the bilateral personalized federated recommendation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0083] Figure 4 : is a structural diagram of an electronic device provided by the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0084] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (also known as random access memory, RAM) 13, etc., wherein the memory stores a computer program that can be executed by the at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12 and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0085] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0086] The processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the bilateral personalized federated recommendation method.

[0087] In some embodiments, the bilateral personalized federated recommendation method may be implemented as a computer program that is tangibly contained in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the bilateral personalized federated recommendation method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the bilateral personalized federated recommendation method in any other appropriate manner (e.g., by means of firmware).

[0088] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0089] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0090] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device or any suitable combination of the foregoing.

[0091] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device that has: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0092] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0093] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0094] In one embodiment, the present invention further includes a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the bilateral personalized federated recommendation method of any embodiment of the present invention.

[0095] The computer program product may be implemented by writing computer program code for performing the operations of the present invention in one or more programming languages, or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0096] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0097] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A bilateral personalized federated recommendation method, characterized in that: include: Determining a target recommendation device based on the information recommendation instruction; wherein the target recommendation device is a user terminal to which item tendency information is recommended; Controlling the target recommendation device to send an information retrieval instruction to a sharing server to retrieve the shared item embedded information stored on the sharing server; Based on the shared item embedding information and the local item embedding information of the target recommendation device, a first recommendation result of the target recommendation device is determined, and the interactive interface of the target recommendation device is controlled to display the first recommendation result so that the user can view the first recommendation result.

2. The method according to claim 1, characterized in that The shared item embedding information is used to represent the item attributes of the item to be recommended, and the local item embedding information is used to represent the user's item preference information; Before determining a first recommendation result for the target recommendation device based on the shared item embedding information and the local item embedding information of the target recommendation device, the method further includes: Processing the shared item embedding information and the local item embedding information using a regularization method so that the shared item embedding information and the local item embedding information are different and complementary; Inducing sparsity of the shared item embedded information to reduce the communication content of the shared item embedded information.

3. The method according to claim 1, characterized in that The determining, based on the shared item embedding information and the local item embedding information of the target recommendation device, a first recommendation result for the target recommendation device includes: Determining an initial item to be recommended and a first weight value of the initial item to be recommended based on the shared item embedding information; Adjusting the first weight value of the initial item to be recommended using the local item embedding information to obtain a second weight value of the initial item to be recommended; The initial items to be recommended are screened using the second weight value and a preset screening condition to obtain the first recommendation result, wherein the preset screening condition is a preset recommendation threshold and / or a preset weight threshold.

4. The method according to claim 2, characterized in that Before determining the target recommended device based on the information recommendation instruction, it also includes: Determining candidate item embedding information, a training device, and a recommendation information prediction model stored on the training device, wherein the candidate item embedding information is general item embedding information stored on the shared server or the latest historical shared item embedding information; Control the training device to retrieve the candidate item embedding information, optimize the recommendation information prediction model using the candidate item embedding information and the sample item embedding information stored on the training device, and determine the shared item embedding information and the local item embedding information of the training device based on the model parameters of the trained recommendation information prediction model when the recommendation information prediction model meets the preset training conditions.

5. The method according to claim 4, characterized in that After controlling the interactive interface of the target recommendation device to display the first recommendation result, the method further includes: Determining an item to be recommended and predicted evaluation information of the item to be recommended based on the first recommendation result; After receiving user evaluation information of the item to be recommended, the model parameters of the recommendation information prediction model are optimized using the user evaluation information and the predicted evaluation information to adjust the shared item embedding information and the local item embedding information of the target recommendation device.

6. The method according to claim 5, characterized in that The optimizing the model parameters of the recommendation information prediction model by utilizing the user evaluation information and the predicted evaluation information includes: determining an error item, type information of the error item, and error data based on difference information between the user evaluation information and the predicted evaluation information; Based on the type information, determine the model parameters to be adjusted of the recommendation information prediction model, and use the error data to adjust the model parameters to be adjusted, wherein a first matching degree between the recommendation information prediction model and the error item after adjusting the model parameters is higher than a second matching degree between the recommendation information prediction model and the error item before adjusting the model parameters.

7. A bilateral personalized federated recommendation device, characterized in that: A bilateral personalized federated recommendation method for implementing any one of claims 1 to 6, wherein the bilateral personalized federated recommendation device comprises: A device determination module, configured to determine a target recommendation device based on the information recommendation instruction; wherein the target recommendation device is a user terminal for which item tendency information is recommended; an information determination module, configured to control the target recommendation device to send an information retrieval instruction to a sharing server, so as to retrieve the shared item embedded information stored on the sharing server; An information recommendation module is used to determine a first recommendation result of the target recommendation device based on the shared item embedding information and the local item embedding information of the target recommendation device, and control the interactive interface of the target recommendation device to display the first recommendation result so that the user can view the first recommendation result.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the bilateral personalized federated recommendation method described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the bilateral personalized federated recommendation method according to any one of claims 1 to 6 when executed.

10. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the bilateral personalized federated recommendation method according to any one of claims 1 to 6.

Images