Information processing model training method, information processing method and related equipment

Abstract

This application relates to the field of artificial intelligence and specifically discloses a training method for an information processing model, an information processing method, and related equipment. The training method includes: obtaining an interactive information sequence of a first user and an interactive information sequence of a second user; generating a local feature of the first user and a global feature of the first user based on a first user feature of the first user; and generating a local feature of the second user based on the first user feature of the second user; calculating a first score based on the local feature of the first user and the global feature of the first user; and calculating a second score based on the global feature of the first user and the local feature of the second user; training a first neural network model, a user feature model, and a discriminant model based on the first score and the second score; and determining an information processing model based on the trained first neural network model and the trained user feature model. This solution can effectively ensure the matching between target information determined in the target domain and users.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a training method for an information processing model, an information processing method, and related equipment. Background Art

[0002] With the development of internet technology, more and more users are sharing and disseminating information through information delivery platforms. After receiving a user's information request, the delivery platform filters the information sent to the user from its information database. In related art, the information delivered by the delivery platform to the user is not highly compatible with the user, resulting in significant discrepancies between the information delivered and the information actually required by the user. This requires the user to interact with the delivery platform multiple times, and the delivery platform to perform multiple information screening and delivery operations before the user can obtain the required information. This requirement for the delivery platform to perform multiple information screening and delivery operations results in low information delivery efficiency. Summary of the Invention

[0003] In view of the above problems, the embodiments of the present application propose a training method for an information processing model, an information processing method and related equipment to solve the problem of low information sending efficiency caused by poor matching between information and users.

[0004] According to one aspect of an embodiment of the present application, a method for training an information processing model is provided, comprising: obtaining an interaction information sequence of a first user and an interaction information sequence of a second user from first user behavior data in a source domain; outputting a first user feature of the first user based on the interaction information sequence of the first user by a first neural network model, and outputting a first user feature of the second user based on the interaction information sequence of the second user by the first neural network model; the first neural network model is pre-trained and determined based on the historical behavior data of the second user in the source domain; generating a local feature of the first user and a global feature of the first user based on the first user feature of the first user by a user feature model; and generating a local feature of the second user based on the first user feature of the second user by the user feature model; calculating a first score based on the local feature of the first user and the global feature of the first user by a discriminant model; and calculating a second score based on the global feature of the first user and the local feature of the second user by the discriminant model; training the first neural network model, the user feature model, and the discriminant model based on the first score and the second score until a training end condition is met; determining an information processing model based on the trained first neural network model and the trained user feature model, the information processing model determining target information to be sent to the user in a target domain.

[0005] According to one aspect of an embodiment of the present application, an information processing method is provided, including: obtaining a candidate information set in a target domain; predicting a target user's behavior label for each information in the candidate information set through the information processing model; the information processing model is trained according to the training method of the information processing model as described above; and determining, in the candidate information set, target information to be sent to the target user based on the behavior label.

[0006] According to one aspect of an embodiment of the present application, a training device for an information processing model is provided, comprising: a first user behavior data acquisition module, configured to acquire an interaction information sequence of a first user and an interaction information sequence of a second user from the first user behavior data of a source domain; a first output module, configured to output a first user feature of the first user according to the interaction information sequence of the first user by a first neural network model, and a second output module, configured to output a first user feature of the second user according to the interaction information sequence of the second user by the first neural network model; the first neural network model is pre-trained and determined based on the historical behavior data of the second user in the source domain; a first generation module, configured to generate a local feature of the first user and a global feature of the first user according to the first user feature of the first user by a user feature model; and a second generation module Block, used for generating the local features of the second user according to the first user features of the second user by the user feature model; a first score calculation module, used for calculating the first score according to the local features of the first user and the global features of the first user by the discriminant model; and a second score calculation module, used for calculating the second score according to the global features of the first user and the local features of the second user by the discriminant model; a first training module, used for training the first neural network model, the user feature model and the discriminant model according to the first score and the second score until the training end condition is met; an information processing model determination module, used for determining the information processing model according to the trained first neural network model and the trained user feature model, the information processing model determining the target information to be sent to the user in the target domain.

[0007] In some embodiments of the present application, the user feature model includes at least two convolutional layers and a pooling layer, and the length of the convolution window corresponding to each of the at least two convolutional layers is different; in this embodiment, the first generation module includes: a first convolution processing unit, which is used to perform convolution processing on the first user feature of the first user according to the corresponding convolution window by each convolution layer in the at least two convolutional layers, so as to obtain multiple intermediate local features of the first user; a first splicing unit, which is used to splice the multiple intermediate local features of the first user to obtain the local features of the first user; a first pooling processing unit, which is used to perform pooling processing on the local features of the first user by the pooling layer, so as to obtain the global features of the first user; in this embodiment, the second generation module includes: a second convolution processing unit, which is used to perform convolution processing on the first user feature of the second user according to the corresponding convolution window by each convolution layer in the at least two convolutional layers, so as to obtain multiple intermediate local features of the second user; a second splicing unit, which is used to splice the multiple intermediate local features of the second user to obtain the local features of the second user.

[0008] In some embodiments of the present application, the first neural network model includes a basic neural network model and an adaptive network inserted into the basic neural network model, and the parameters of the basic neural network model are pre-trained and determined based on the historical behavior data of the second user in the source domain; the training device of the information processing model also includes: a parameter fixing module for fixing the parameters of the basic neural network model; in this embodiment, the first training module is further configured to: adjust the parameters of the adaptive network, the user feature model and the discriminant model according to the first score and the second score and continue training until the training end condition is reached.

[0009] In some embodiments of the present application, the basic neural network model includes an embedding layer and multiple cascaded converter neural networks, the first neural network model is divided into N cascaded sub-neural networks, the sub-neural networks include a converter neural network and at least one adaptive network inserted into the converter neural network; N is an integer greater than 1; the first output module includes: an input information acquisition unit for acquiring input information corresponding to the k-level sub-neural network; wherein, if k=1, the input information corresponding to the k-level sub-neural network is the embedding vector output by the embedding layer according to the interaction information sequence of the first user; if k>1, the k-level sub-neural network is the embedding vector output by the embedding layer according to the interaction information sequence of the first user; The input information corresponding to the sub-neural network is the k-1-level sub-neural network outputting the k-1-level intermediate user features of the first user; 0<k≤N, k is an integer; the first input unit is used to input the input information corresponding to the k-level sub-neural network into the k-level sub-neural network; the feature extraction unit is used to perform feature extraction by the k-level sub-neural network according to the corresponding input information, and output the k-level intermediate user features of the first user; wherein, if k=N, the k-level intermediate user features of the first user are used as the first user features of the first user; if k<N, the k-level intermediate user features of the first user are used as the input information of the k+1-level sub-neural network.

[0010] In some embodiments of the present application, the converter neural network includes a cascaded multi-head attention layer, a first feedforward neural network layer, a first summation and normalization layer, a second feedforward neural network layer, and a second summation and normalization layer; each of the sub-neural networks includes an adaptive network inserted between the first feedforward neural network layer and the first summation and normalization layer, and an adaptive network inserted between the second feedforward neural network layer and the second summation and normalization layer; the sub-neural network also includes an identity mapping from the input of the multi-head attention layer to the input of the first summation and normalization layer, and an identity mapping from the input of the second feedforward neural network layer to the input of the second summation and normalization layer.

[0011] In some embodiments of the present application, the adaptive network includes a cascaded first fully connected layer, an intermediate fully connected layer, an activation layer and a last fully connected layer, the number of neurons in the first fully connected layer is equal to the number of neurons in the last fully connected layer, and the number of neurons in the first fully connected layer is greater than the number of neurons in the intermediate fully connected layer; the adaptive network also includes an identity mapping from the input of the first fully connected layer to the output of the last fully connected layer.

[0012] In some embodiments of the present application, the training device of the information processing model also includes: a target node selection module, which is used to select multiple target nodes in a reference interaction information sequence; the reference interaction information sequence refers to the interaction information sequence of each user in the second user historical behavior data; a replacement module, which is used to block the multiple target nodes in the reference interaction information sequence to obtain a sample interaction information sequence; a sample interaction information sequence input module, which is used to input the sample interaction information sequence into the basic neural network model; a prediction information output module, which is used for the basic neural network model to output prediction information corresponding to the target node according to the sample interaction information sequence; a prediction error determination module, which is used to determine the prediction error based on the prediction information and the target node; a second training module, which is used to adjust the parameters of the basic neural network model according to the prediction error and continue training until the pre-training end condition is reached.

[0013] According to one aspect of an embodiment of the present application, an information processing device is provided, comprising: a candidate information set acquisition module for acquiring a candidate information set in a target domain; a behavior label prediction module for predicting the behavior label of a target user for each information in the candidate information set through the information processing model; the information processing model is trained according to the training method of the information processing model as described above; and a target information determination module for determining, in the candidate information set, target information to be sent to the target user based on the behavior label.

[0014] In some embodiments of the present application, the information processing model includes a trained first neural network model, a trained user feature model, an embedding search layer and a classification layer; the behavior label prediction module includes: a global feature search unit, used to obtain the global features of the target user in the user global feature set; the global features in the user global feature set are generated by the trained first neural network model and the trained user feature model according to the first user behavior data; an embedding vector generation unit, used to generate the embedding vector of each information in the candidate information set by the embedding search layer; a behavior label output unit, used to output the behavior label of the target user for each information in the candidate information set by the classification layer based on the global features of the target user and the embedding vector of each information in the candidate information set.

[0015] In some embodiments of the present application, the information processing device further includes: a receiving module for receiving an information request, wherein the information request indicates a user identifier of the target user; and a sending module for sending the target information to the target user according to the user identifier of the target user.

[0016] In some embodiments of the present application, the information processing device also includes: an interactive behavior information acquisition module, used to obtain the interactive behavior information of the target user with respect to the target information; and an update training module, used to update and train the information processing model based on the interactive behavior information of the target user with respect to the target information.

[0017] According to one aspect of an embodiment of the present application, an electronic device is provided, comprising: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the training method of the information processing model or the information processing method as described above is implemented.

[0018] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the training method of the information processing model or the information processing method as described above is implemented.

[0019] According to one aspect of an embodiment of the present application, a computer program product is provided, comprising computer instructions, which, when executed by a processor, implement the training method of the information processing model or the information processing method as described above.

[0020] In the solution of the present application, a first neural network model is pre-trained using second user behavior data from a source domain, so that the first neural network model learns the user's user characteristics based on the user's interaction information sequence in the source domain. On this basis, the first neural network model and the user characteristic model are combined with a discriminant model and subjected to unsupervised comparative learning using the first user behavior data from the source domain, thereby widening the differences between the characteristics of different users and improving the accuracy of the user characteristics extracted based on the interaction information sequence. Thus, after determining the information processing model based on the trained first neural network model and the user characteristic model, because the information processing model accurately learns the user's characteristics in the source domain, and the user's characteristics reflect the user's information preferences, the information processing model determines the target information to be sent to the user in the target domain based on the user characteristics learned in the source domain. Because the target information determination process incorporates the user's user characteristics in the source domain, the matching between the target information and the user can be improved, thereby effectively reducing the number of interactions with the user, effectively solving the problem of low information transmission efficiency caused by poor matching between information and users in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification, are used to explain the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0022] Figure 1 A schematic diagram of an information sending system provided by an exemplary embodiment of the present application is shown.

[0023] Figure 2 It is a flowchart of a method for training an information processing model according to an embodiment of the present application.

[0024] Figure 3 3 is a schematic diagram illustrating unsupervised contrastive learning during the fine-tuning training phase according to an embodiment of the present application.

[0025] Figure 4 is a schematic diagram of a basic neural network model according to an embodiment of the present application.

[0026] Figure 5 Schematic diagram of a sub-neural network according to an embodiment of the present application.

[0027] Figure 6 1 is a schematic diagram of the structure of an adaptive network according to an embodiment of the present application.

[0028] Figure 7 This is a flowchart of the steps before step 220 according to an embodiment of the present application.

[0029] Figure 8 It is a flowchart of an information processing method according to an embodiment of the present application.

[0030] Figure 9 FIG. 8 is a flowchart of step 820 according to an embodiment of the present application.

[0031] Figure 10 2 is a schematic diagram of predicting behavior labels using an information processing model according to an embodiment of the present application.

[0032] Figure 11 This is a schematic diagram showing displaying target information in a user interface according to an embodiment of the present application.

[0033] Figure 12 FIG. 4 is a block diagram of a training apparatus for an information processing model according to an embodiment.

[0034] Figure 13 is a block diagram of an information processing device according to an embodiment of the present application.

[0035] Figure 14 A schematic diagram of the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0036] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0037] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner.In the following description, many specific details are provided so as to provide a full understanding of the embodiments of the present application. However, it will be appreciated by those skilled in the art that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps etc. can be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0038] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0039] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, while others may be combined or partially combined. Therefore, the actual execution order may vary depending on the actual situation.

[0040] It should be noted that the term "plurality" used in this document refers to two or more. "And / or" describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. The character " / " generally indicates an "or" relationship between the associated objects.

[0041] Before describing the present application in detail, the following explanations are given for the terms used in the present application:

[0042] Source domain: In transfer learning, the source domain refers to the field to which the existing knowledge belongs.

[0043] Target domain: also known as the target field, refers to the field where new knowledge is to be learned, which can also be understood as the field where the current target task needs to be processed.

[0044] In the information transmission scenario, the domain to which an application already has a sufficient amount of user interaction data belongs can be called the source domain. The domain to which an application already has no or less user interaction data belongs can be called the target domain. For example, if a first application already has a large amount of user interaction data for messages sent within the first application, the domain to which the first application belongs can be called the source domain. However, if a second application lacks user interaction data, the domain to which the second application belongs can be called the target domain.

[0045] In order to solve the problem of low information sending efficiency due to poor matching between information and users in the existing technology, the solution of the present application is proposed. An information processing model is constructed and trained based on artificial intelligence technology, and the user interaction information sequence of users for information in the source domain is used to train the information processing model, and the user characteristics of each user in the source domain are learned. On this basis, the user characteristics learned by the information processing model in the source domain are used to specifically determine the information to be sent to the user in the target domain, thereby ensuring the matching between the target information determined for the user in the target domain and the user, thereby improving the efficiency of information sending.

[0046] In some application scenarios, such as when new applications lack user interaction data (this scenario can be called a cold start scenario, and the relevant users are called cold users), there may be a situation where the information sent to the user does not match the user well, resulting in the need to send information to the user multiple times. Therefore, the solution of the present application can also be used to determine the target information to be sent to the user in the cold start scenario.

[0047] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0048] Artificial intelligence technology is an interdisciplinary subject that covers a wide range of fields, including both hardware-level and software-level technologies. Basic artificial intelligence technologies generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operating / interactive systems, mechatronics, and other technologies. Artificial intelligence software technologies mainly include computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning. The solution of this application mainly uses natural language processing technology to mine user features based on the information that triggers user behavior in the source domain to obtain the user's user features. The user features of the user in the source domain are applied across domains to the target domain to achieve information recommendation across neighboring domains.

[0049] In some embodiments of the present application, the target domain may be a field or product field similar to that in the source domain. For example, if the information to be recommended in the source domain and the target domain is news, the target domain and the source domain can be regarded as similar fields; for another example, if the information to be recommended in the source domain and the target domain is aggregated information such as blogs, public account articles, videos, etc., the target domain and the source domain can be regarded as similar fields. In some embodiments, if the type of information in the target domain is the same as the type of information in the source domain, the target domain can be regarded as a similar field to the source domain. The types of information include articles (blogs, news, public account articles, etc.), short videos, news, advertisements, product links, etc. Correspondingly, the information to be recommended in the source domain and the target domain can be articles, short videos, news, advertisements, etc.

[0050] Figure 1 FIG. 1 is a schematic diagram of an information sending system provided by an exemplary embodiment of the present application. The information sending system includes a terminal 110 and a sending platform 120 .

[0051] The sending platform 120 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0052] The sending platform 120 is used to provide background services for applications that support information sending. Optionally, the sending platform 120 undertakes the main computing work and the terminal 110 undertakes the secondary computing work; or, the sending platform 120 and the terminal 110 adopt a distributed computing architecture to perform collaborative computing.

[0053] Optionally, the sending platform 120 includes a first sending server 121 and a second sending server 122. The first sending server 121 is used to provide services related to information sending in the source domain, such as determining the information in the source domain to be sent to each user, collecting user interaction data for the information sent in the source domain, etc.

[0054] The second sending server 122 is used to provide services related to information transmission in the target domain, such as determining the target information in the target domain to be sent to each user. In some embodiments, the second sending server 122 is also in communication with the first sending server 121. An information processing model is deployed on the second sending server 122. This information processing model can be trained based on information in the source domain sent to users by the first sending server 121, as well as user interaction data regarding the sent information in the source domain. The trained information processing model can then be used to determine the target information in the target domain to be sent to users.

[0055] Terminal 110 is connected to the sending platform 120 via a wireless network or a wired network. Terminal 110 can be an electronic device such as a smartphone, laptop, game console, desktop computer, tablet computer, e-book reader, smart speaker, etc. Terminal 110 has an application installed and running that supports information sending. This application can be any of a news application, an instant messaging application, a reading application, a social application that aggregates people based on topics, channels, or circles, a shopping-based social application, a browser application, or a video application. Illustratively, terminal 110 is used by a first user, and the application running on terminal 110 has the first user account logged in.

[0056] The applications running on terminal 110 that support information transmission can be two applications: one application is a first application corresponding to the source domain, and the other application is a second application corresponding to the target domain. Based on this, the first application can send the first information in the source domain to the user; the second application can recommend information in the target domain to the user. For example, the first application can be an instant messaging application, and the information in the source domain can be articles, news, blogs, etc. from a public account; the second application can be a reading application, and the information to be sent in the target domain can be blogs, book reviews, news, etc. within the reading application.

[0057] When the first application can be run in the terminal 110, when the terminal 110 is running the first application, the first sending server 121 can send information in the source domain to the user where the first application is located, and then the first application detects the interaction behavior triggered by the user for the sent information in the source domain, and reports the user's interaction data for the sent information in the source domain to the first sending server 121, so as to use the reported user's interaction data for the sent information in the source domain as training data for the information processing model.

[0058] The first sending server 121 can send the interaction data of users regarding the recommended information in the source domain reported by numerous terminals 110 running the first application to the second sending server 122, and the second sending server 122 trains the information processing model according to the method of the present application, and uses the information processing model to specifically determine the target information in the target domain for the users in the second application, and sends the determined target information in the target domain to the users in the second application.

[0059] When the second application can still be run in the terminal 110, the second sending server 122 can determine the target information to be sent to the user in the target domain for the user where the second application is located according to the information processing method of this application, and send the target information in the target domain to the terminal 110.

[0060] It is understandable that the terminal 110 may run only the first application or the second application, or may run both the first application and the second application. Figure 1 The number of terminals in the information sending system shown in FIG can be greater, and the information sending system can also include other terminals.

[0061] The following is a detailed description of the implementation details of the technical solution of the embodiment of the present application:

[0062] Figure 2 The flowchart of the training method of the information processing model according to one embodiment of the present application is shown. The method can be executed by a computer device with processing capabilities, such as a server, etc., which is not specifically limited here. Figure 2 As shown, the method includes at least steps 210 to 280, which are described in detail as follows:

[0063] Step 210: Acquire the interaction information sequence of the first user and the interaction information sequence of the second user from the first user behavior data in the source domain.

[0064] An information set is provided in the source domain, and based on the information set, information in the source domain can be sent to the user in a targeted manner. After the information in the information set is sent to the user, the user's interaction data on the information sent in the source domain is collected. The user's interaction data on the recommended information can be reflected by the user's interactive behavior indicating interest in the information triggered by the recommended information. For ease of description, the interactive behavior indicating the user's interest in the information is referred to as interested user behavior. Interested user behavior includes, for example, click behavior, collection behavior, like behavior, comment behavior, etc. triggered on the recommended information. In some embodiments, only after the user clicks to enter the details page of the recommended information can the user trigger behaviors such as likes, collections, or comments on the recommended information. Alternatively, only interactive data reflecting the user's click behavior triggered on the recommended information can be collected.

[0065] Collect interaction data of multiple users on sent information in the source domain for a large number of users. Based on the interaction data of many users on the sent information, arrange multiple sent information that trigger interactive behaviors of the same user in chronological order to obtain an interaction information sequence for each user. In other words, the user's interaction information sequence refers to an information sequence obtained by arranging the sent information that triggers user behaviors of interest to the user in chronological order. The chronological order can be determined based on the time when the user triggers user behaviors of interest to the sent information. It is understandable that in the user's interaction information sequence, the sent information can be represented by the information identifier of the sent information, such as the information ID. For example, if the information sent to the user in the source domain includes A, B, C, D, E, F; wherein the information clicked by the user in sequence includes A, C, E, F, then the user's interaction information sequence can be<A,C,E,F> .

[0066] Furthermore, to indicate the domain to which the interaction information sequence belongs and the user to which the interaction information sequence corresponds, a triplet array can be formed by combining the user ID, the domain ID of the domain to which the interaction information sequence belongs, and the interaction information sequence. Thus, the first user behavior data of the source domain is a collection of multiple triplet arrays.

[0067] The first user behavior data in the source domain includes interaction information sequences of multiple users in the source domain. In this solution, since the information processing model is fine-tuned and trained through unsupervised contrastive learning, the interaction information sequences of two different users in the source domain are obtained from the first user behavior data in the source domain each time. To facilitate the distinction between the two users, one of the users is called the first user and the other user is called the second user.

[0068] In step 220, the first neural network model outputs the first user feature of the first user based on the interaction information sequence of the first user, and in step 230, the first neural network model outputs the first user feature of the second user based on the interaction information sequence of the second user; the first neural network model is pre-trained and determined based on the historical behavior data of the second user in the source domain.

[0069] The solution of this application is to apply the knowledge learned in the source domain to the information sending task in the target domain through transfer learning. Therefore, the information processing model is trained with the help of user interaction information in the source domain so that the information processing model can be used to determine the target information to be sent in the target domain.

[0070] In the solution of the present application, the information processing model is trained through two training stages, which can be called the pre-training stage and the fine-tuning training stage. Training is performed first in the pre-training stage and then in the fine-tuning training stage. Therefore, the training data in the source domain is divided into two parts, one of which is used in the pre-training stage and the other is used in the fine-tuning training stage. In the solution of the present application, the training data used in the pre-training stage is referred to as the second user behavior data, and the training data used in the fine-tuning training stage is referred to as the first user behavior data.

[0071] It is understandable that even if the same information is sent to different users, due to the differences in the information that different users are interested in or prefer, the interactive behaviors triggered by different users for the same information will also be different. In other words, the interactive behaviors triggered by users for the information sent in the source domain reflect, to a certain extent, the user's preference characteristics for information. Based on this, in the solution of the present application, in the pre-training stage, the first neural network model is trained by the second user behavior data in the source domain, so that the first neural network model has the ability to output the user's feature representation according to the user's interactive information sequence. For the sake of convenience, the feature representation of the user output by the first neural network model is referred to as the first user feature.

[0072] The first neural network model may be a model composed of a neural network, such as a CNN (Convolutional Neural Network), an RNN (Recurrent Neural Network), an LSTM (Long Short-Term Memory Neural Network), a fully connected neural network, or a Transformer model.

[0073] After pre-training the first neural network model, since the first neural network model has the ability to output the user's feature representation according to the user's interaction information sequence, the first neural network model can output the first user feature of the first user according to the interaction information sequence of the first user in the source domain, and output the first user feature of the second user according to the interaction information sequence of the second user in the source domain.

[0074] In step 240 , the user feature model generates a local feature of the first user and a global feature of the first user based on the first user feature of the first user; and in step 250 , the user feature model generates a local feature of the second user based on the first user feature of the second user.

[0075] The user feature model is used to extract deep features based on the user's first user feature. The user feature model can be a neural network model constructed by a neural network such as a convolutional neural network, a pooling neural network, or a fully connected neural network.

[0076] The features that reflect the local part of the user output by the user feature model are called local features, and the features that reflect the global part of the user output by the user feature model are called global features. Among them, the global features of the user are obtained by further processing the local features of the user.

[0077] In some embodiments of the present application, the user feature model includes at least two convolutional layers and a pooling layer, and the length of the convolution window corresponding to each of the at least two convolutional layers is different; in this embodiment, the user feature model can output the local features and global features of the first user through the following process: each of the at least two convolutional layers performs convolution processing on the first user features of the first user according to the corresponding convolution window to obtain multiple intermediate local features of the first user; the multiple intermediate local features of the first user are spliced ​​to obtain the local features of the first user; the pooling layer performs pooling processing on the local features of the first user to obtain the global features of the first user.

[0078] Similarly, the user feature model can output the local features of the second user through the following process: each convolution layer in at least two convolution layers performs convolution processing on the first user features of the second user according to the corresponding convolution window to obtain multiple intermediate local features of the second user; the multiple intermediate local features of the second user are spliced ​​to obtain the local features of the second user.

[0079] For a convolution kernel, different convolution window lengths result in different receptive fields of view. Therefore, convolution layers with different convolution windows are used to perform convolution processing on the first user's first user feature. The resulting multiple intermediate local features of the first user are extracted from the first user feature under different receptive fields of view. Each convolution layer outputs an intermediate local feature of the user.

[0080] In this embodiment, after obtaining multiple intermediate local features of a user (a first user and a second user), the multiple intermediate local features are concatenated and the concatenated features are used as the user's local features. Based on this, the concatenated local features of the user are input into a pooling layer, which performs pooling on the user's local features and outputs the user's global features. The pooling process performed may be average pooling, maximum pooling, or the like.

[0081] Step 260 : The discriminant model calculates a first score based on the local features of the first user and the global features of the first user; and the discriminant model calculates a second score based on the global features of the first user and the local features of the second user.

[0082] The discriminant model can be a model constructed by a neural network such as a convolutional neural network and a fully connected network. The discriminant model calculates a first score based on the local features of the first user and the global features of the first user, and the discriminant model calculates a second score based on the global features of the first user and the local features of the second user. Based on the calculated scores, the parameters of the model (the first neural network model, the user feature model, and the discriminant model) are reversely adjusted to narrow the gap between the local features and the global features of the same user, and widen the gap between the global features of the first user and the local features of the second user, thereby maximizing the amount of information between the global features and local features of the users, and widening the difference between the features of different users.

[0083] As can be seen from the above, in the solution of the present application, the user feature model is not only used to extract features from the first user's features, but also used to construct negative samples for unsupervised contrastive learning. Specifically, in this solution, the global features of the first user and the local features of the first user are used as positive samples, and the local features of the second user are used as negative samples. Since negative samples for contrastive learning are automatically constructed through the user feature model, there is no need to label the data during the fine-tuning training phase, which greatly reduces the workload of model training. Moreover, since no labeling is required, it can avoid the situation where labeling errors affect the model training effect.

[0084] In some embodiments, the first score reflects the similarity between the first user's local features and their global features, while the second score reflects the similarity between the first user's global features and the second user's local features. Therefore, during training, it is desirable to have a higher first score and a lower second score, thereby increasing the differentiation between the features of different users.

[0085] Step 270 : Training the first neural network model, the user feature model, and the discriminant model according to the first score and the second score until a training end condition is met.

[0086] In some embodiments of the present application, loss functions can be set for the first neural network model, the user feature model, and the discriminant model respectively, and a target loss function can be set based on the loss function of the first neural network model, the loss function of the user feature model, and the loss function of the discriminant model. On this basis, the loss functions of the first neural network model, the user feature model, and the discriminant model are calculated respectively according to the first score and the second score, and then the function value of the target loss function is determined, and the parameters of at least one model among the first neural network model, the user feature model, and the discriminant model are adjusted according to the function value of the calculated target loss function, and the above steps 220-260 are performed again by adjusting the parameters of the first neural network model, the user feature model, and the discriminant model until the function value of the calculated target loss function makes the target function converge, and then the interaction information sequence of the other two users is obtained from the first user behavior data to repeat the above steps 210-270 until the training end condition is met.

[0087] In some embodiments, when adjusting the parameters of the first neural network model, the user feature model, and the discriminant model based on the function value of the target loss function, the parameters of each model can be adjusted in stages, with only the parameters of one model adjusted in one stage, and the parameters of the other models adjusted in the next stage. For example, if in one stage, the parameters of the user feature model and the discriminant model are fixed, and only the parameters of the first neural network model are adjusted; in the next stage, the parameters of the first neural network model and the discriminant model are fixed, and the parameters of the user feature model are adjusted; in the next stage, the parameters of the first neural network model and the user feature model are fixed, and the parameters of the discriminant model are adjusted; thus, in the fine-tuning training stage, the parameters of each model are adjusted alternately according to the above process. Of course, in other embodiments, it is also possible to set the parameters of only two models to be adjusted in one stage, which is not specifically limited here. Since only the parameters of one or two models are adjusted in one parameter adjustment process, the amount of parameter adjustment is reduced, and the training speed and efficiency of the model are improved.

[0088] In some embodiments, in order to maximize the information between the global features and local features of a user and to differentiate between features of different users, the loss function set for the discriminant model may be:

[0089] Loss1=log(1+exp -t1 )+log(1+exp t2 );

[0090] Among them, Loss1 is the function value of the loss function of the discriminant model, t1 represents the first score calculated by the discriminant model, and t2 represents the second score calculated by the discriminant model.

[0091] Step 280: determining an information processing model based on the trained first neural network model and the trained user feature model; the information processing model determines target information to be sent to the user in the target domain.

[0092] In this solution, the discriminant model can be a model used to assist in fine-tuning training. Therefore, the information processing model can be determined by combining the trained first neural network model and the trained user feature model. In some embodiments, due to the difference between the tasks in the fine-tuning training stage and the recommendation tasks in the target domain, that is, the task in the fine-tuning training stage is to output user features based on the user's interactive information sequence, while the recommendation task in the target domain is to predict the user's behavior label for the information in the target domain, and the behavior label indicates the predicted user behavior triggered by the information. Therefore, a classification layer can be added on the basis of the trained first neural network model and the trained user feature model, and the classification layer is used to output the predicted user behavior label for the information in the target domain.

[0093] The target user generally refers to a user in the target domain to whom information is to be sent. In some embodiments, the target user may be a new user in the target domain, or a user whose interaction data in the target domain is less than a preset threshold.

[0094] In the target domain, due to the lack of user interaction data for information in the target domain, or the small amount of user interaction data for information in the target domain, in this case, since the information processing model trained by the first user behavior data and the second user behavior data in the source domain accurately learns the user's characteristics, the user characteristics learned by the information processing model can be transferred to the information sending task in the target domain. The information processing model applies the user characteristics learned in the source domain to predict the user's behavior label for information in the target domain, and then determines the target information to be sent to the user in a targeted manner based on the behavior label.

[0095] In the solution of the present application, a first neural network model is pre-trained using second user behavior data from a source domain, so that the first neural network model learns the user's user characteristics based on the user's interaction information sequence in the source domain. On this basis, the first neural network model and the user characteristic model are combined with a discriminant model and subjected to unsupervised comparative learning using the first user behavior data from the source domain, thereby widening the differences between the characteristics of different users and improving the accuracy of the user characteristics extracted based on the interaction information sequence. Thus, after determining the information processing model based on the trained first neural network model and the user characteristic model, because the information processing model accurately learns the user's characteristics in the source domain, and the user's characteristics reflect the user's information preferences, the information processing model determines the target information to be sent to the user in the target domain based on the user characteristics learned in the source domain. Because the determination of the target information incorporates the user's user characteristics in the source domain, the matching between the target information and the user can be improved, thereby effectively reducing the number of interactions with the user, effectively solving the problem of low information transmission efficiency caused by poor matching between information and users in the prior art.

[0096] Moreover, the solution of the present application can also be used to determine the target information to be sent to the user in a cold start scenario, which can effectively improve the matching between the target information determined for the user and the user in the cold start scenario of the target domain, without the need to send information to the user multiple times, thereby improving the efficiency of information sending. For users, it improves the user acquisition efficiency and enhances the user experience.

[0097] Figure 3 FIG. 1 is a schematic diagram showing unsupervised contrastive learning in fine-tuning training according to an embodiment of the present application. Figure 3 As shown, the pre-trained first neural network model 310 outputs the first user feature 341 of the first user and the first user feature 342 of the second user respectively. Then, the user feature model 320 generates local features and global features for the first user, and generates local features for the second user. Specifically, Figure 3 As shown, convolutional layers with different window lengths in the user feature model perform N-gram local feature extraction on first user feature 341 of the first user to obtain multiple intermediate local features 351 of the first user. These multiple intermediate local features 351 of the first user are then concatenated to obtain local features 361 of the first user. Subsequently, local features 361 of the first user are average pooled to obtain global features 362 of the first user. Following a similar process, user feature model 320 outputs intermediate local features 352 and local features 363 of the second user.

[0098] Subsequently, the discriminant model 330 calculates a first score based on the first user's local features 361 and the first user's global features 362, and the discriminant model 330 calculates a second score based on the second user's local features 363 and the first user's global features 362. The parameters of the first neural network model 310, the user feature model 320, and the discriminant model 330 are then adjusted inversely based on the calculated first and second scores.

[0099] In some embodiments of the present application, the first neural network model includes a basic neural network model and an adaptive network inserted into the basic neural network model, and the parameters of the basic neural network model are determined by pre-training based on the second training data; before step 220, the method also includes: fixing the parameters of the basic neural network model; in this embodiment, step 270 includes: adjusting the parameters of the adaptive network, the user feature model and the discriminant model based on the first score and the second score and continuing training until the training end condition is reached.

[0100] In this solution, the parameters of the basic neural network model are adjusted during the pre-training process; after pre-training, the parameters of the basic neural network model are fixed, and an adaptive network is inserted into the basic neural network model. Therefore, in the fine-tuning training stage, for the first neural network model, only the parameters of the adaptive network are adjusted, while the parameters of the basic neural network model remain unchanged. The learning results of the pre-training stage can be retained in the fine-tuning training stage, while the amount of parameter adjustment is greatly reduced, effectively improving the model training speed and training effect.

[0101] In some embodiments of the present application, the basic neural network model includes an embedding layer and a plurality of cascaded transformer neural networks. Figure 4 is a schematic diagram of a basic neural network model according to an embodiment of the present application. Figure 4 The embedding layer of the basic neural network model 400 is not shown in FIG. Only the multiple cascaded transformer neural networks 410 in the basic neural network model are shown. The embedding layer is used to transform information to obtain the embedding vector of each information. For example, if the input is a sequence of interactive information of a user in the source domain, the embedding layer can generate a corresponding embedding vector for each information ID corresponding to the user interactive information sequence, such as Figure 4 As shown, for each message ID's corresponding embedding vector, the message ID is positionally encoded based on its position in the user interaction message sequence. The positional encoding and embedding vector corresponding to the message ID are then superimposed to form the embedding vector for the message ID. The embedding vectors for all message IDs in the user interaction message sequence are then input into the primary transformer neural network 410 for feature extraction, and the output features are then input into the next-level transformer neural network 410.

[0102] The Transformer neural network, also known as the Transformer model, is a model structure used to encode words and sentences in the field of NLP (Natural Language Processing) research. It is an attention-based model that can solve the problem of long-distance sequence dependencies. It does not have the severe gradient diffusion and fixed structural limitations of recurrent neural networks, nor does it need to consider the range of the receptive field like convolutional neural networks. In general, the Transformer model not only has good implementation effects, but also can be parallelized.

[0103] In some embodiments of the present application, the adaptive network can be inserted into the converter network. Of course, the adaptive network can be inserted into each level of the converter neural network, or can be inserted at intervals in multiple cascaded converter neural networks. In other embodiments, the adaptive network can also be inserted between two adjacent levels of converter networks.

[0104] In some embodiments of the present application, the first neural network model is divided into N cascaded sub-neural networks, and the sub-neural networks include a converter neural network and at least one adaptive network inserted into the converter neural network; N is an integer greater than 1; in this embodiment, step 220 includes: obtaining input information corresponding to the k-level sub-neural network; wherein, if k=1, the input information corresponding to the k-level sub-neural network is the embedding vector output by the embedding layer according to the interaction information sequence of the first user; if k>1, the input information corresponding to the k-level sub-neural network is the k-1-level intermediate user features of the first user output by the k-1-level sub-neural network; 0<k≤N, k is an integer; the input information corresponding to the k-level sub-neural network is input into the k-level sub-neural network; the k-level sub-neural network performs feature extraction based on the corresponding input information and outputs the k-level intermediate user features of the first user; wherein, if k=N, the k-level intermediate user features of the first user are used as the first user features of the first user; if k<N, the k-level intermediate user features of the first user are used as the input information of the k+1-level sub-neural network.

[0105] A multi-level neural network is used to extract features based on the user's interaction information sequence in the source domain, ensuring the accuracy of the extracted first user features. It is understood that the process for extracting the first user features of the second user is similar to that of the first user, and the extraction process of the first user features of the second user is not further described here.

[0106] Figure 5 is a schematic diagram of a sub-neural network according to an embodiment of the present application, such as Figure 5As shown, the sub-neural network 500 includes a converter neural network and two adaptive networks inserted into the converter neural network. Figure 5 As shown, the converter neural network includes a cascaded multi-head attention layer 510, a first feedforward neural network layer 520, a first summation and normalization layer 530, a second feedforward neural network layer 540, and a second summation and normalization layer 550; each sub-neural network 500 includes an adaptive network 560 inserted between the first feedforward neural network layer 520 and the first summation and normalization layer 530, and an adaptive network 560 inserted between the second feedforward neural network layer 540 and the second summation and normalization layer 550; the sub-neural network 500 also includes an identity mapping from the input of the multi-head attention layer 510 to the input of the first summation and normalization layer 530, and an identity mapping from the input of the second feedforward neural network layer 540 to the input of the second summation and normalization layer 550.

[0107] Generally speaking, the deeper the network layer, the more features of different levels can be extracted, and the more combinations of features of different levels can be obtained. However, as the number of network layers continues to deepen, problems such as gradient vanishing and gradient explosion are very likely to occur, which will lead to the degradation of network performance. In order to solve this problem, residual connections are introduced in this solution, such as Figure 5 In the

[15] , the input of the multi-head attention layer points to the identity mapping of the input of the first summation and normalization layer, and the input of the second feedforward neural network layer points to the identity mapping of the input of the second summation and normalization layer, thereby applying shallow information to deep calculations to avoid the problems of gradient vanishing and gradient exploding.

[0108] Figure 6 is a schematic diagram of the structure of an adaptive network according to an embodiment of the present application, such as Figure 6 As shown, the adaptive network (Adapter) 560 includes a cascaded first fully connected layer 610, an intermediate fully connected layer 620, an activation layer 630 and a last fully connected layer 640. The number of neurons in the first fully connected layer 610 is equal to the number of neurons in the last fully connected layer 640, and the number of neurons in the first fully connected layer 610 is greater than the number of neurons in the intermediate fully connected layer 620; the adaptive network 560 also includes an identity mapping from the input of the first fully connected layer 610 to the output of the last fully connected layer 640.

[0109] It is worth mentioning that in this embodiment, according to the data transmission direction in the adaptive network, the fully connected layers in the adaptive network (Adapter) are divided into the first fully connected layer, the middle fully connected layer and the last fully connected layer.

[0110] For a fully connected layer, the number of neurons in the fully connected layer determines the dimension of the vector output by the fully connected layer. That is, if the number of neurons in the fully connected layer is M, then the vector output by the fully connected layer is M-dimensional. Therefore, in this scheme, the number of neurons in the first fully connected layer is the same as the number of neurons in the last fully connected layer, while the number of neurons in the intermediate fully connected layers is smaller than the number of neurons in the first fully connected layer, resulting in the adaptive network having a bottleneck-style MLP (multilayer perceptron) structure.

[0111] The activation function is configured in the activation layer 630, and the output of the previous layer is nonlinearly transformed through the activation function. The activation function can be a ReLU function, a nonlinear activation function based on Gaussian Error Linear Units (GELU), a hyperbolic tangent function (Tanh function), etc., which are not specifically limited here. Since the fully connected layer only performs a linear transformation (affine transformation) on the data, and the superposition of multiple linear transformations is still a linear transformation, the nonlinearity of the adaptive network is increased by the activation layer. It is worth mentioning that Figure 6 Only one intermediate fully connected layer 620 in the adaptive network is exemplarily shown. In other embodiments, the adaptive network may include multiple intermediate fully connected layers.

[0112] Similarly, in the adaptive network, an identity mapping is used from the input of the first fully connected layer 610 to the output of the last fully connected layer 640 to avoid problems such as gradient vanishing and gradient explosion caused by the increase in the number of network layers in the adaptive network. The input of the shallow first fully connected layer is introduced into the deep last fully connected layer for calculation.

[0113] In some embodiments of the present application, before step 220, as Figure 7 As shown, the method further includes:

[0114] Step 710 : Select multiple target nodes in a reference interaction information sequence; the reference interaction information sequence refers to an interaction information sequence of each user in the second user's historical behavior data.

[0115] As described above, the interactive information sequence is formed by sequentially sorting multiple information. One information in the interactive information sequence can be regarded as a node. Therefore, the reference interactive information sequence includes multiple nodes, and then one or more nodes in the reference interactive information sequence are selected as target nodes.

[0116] Of course, in other embodiments, a predetermined number or a predetermined ratio may be pre-set, and target nodes may be selected from the reference interaction information sequence according to the predetermined number or ratio. The predetermined number indicates the number of target nodes, and the predetermined ratio indicates the ratio of the number of target nodes in the reference interaction information sequence. In a specific embodiment, nodes may be randomly selected as target nodes.

[0117] Step 720: mask multiple target nodes in the reference interaction information sequence to obtain a sample interaction information sequence.

[0118] As described above, each node in the reference interaction information sequence indicates information about an interaction behavior triggered by the corresponding user, such as information about a click behavior triggered by the user. In this embodiment, the target node is blocked, allowing the basic neural network model to learn the information indicated by the blocked target node. The sample interaction information sequence refers to the information sequence after the target node in the reference interaction information sequence is blocked.

[0119] Step 730: Input the sample interaction information sequence into the basic neural network model.

[0120] In step 740, the basic neural network model outputs prediction information corresponding to the target node based on the sample interaction information sequence.

[0121] As described above, the basic neural network model needs to output the information indicated by the occluded target node in the sample interaction information sequence based on the sample interaction information sequence. Specifically, the basic neural network model will make predictions based on the information indicated by the adjacent nodes before and after the occluded node in the sample interaction information sequence. In other words, the basic neural network model learns the contextual relationship of the nodes in the sample interaction information sequence and uses the unoccluded nodes to predict the occluded information of the target node to obtain the prediction information corresponding to each target node. The prediction information corresponding to each target node is used to indicate the information indicated by the occluded target node predicted by the basic neural network model.

[0122] Step 750: Determine the prediction error based on the prediction information and the target node.

[0123] In the reference interaction information sequence, the information indicated by the target node is clear. Therefore, the prediction error can be determined based on the information indicated by the target node in the reference interaction information sequence and the prediction information corresponding to the target node. In some embodiments, a loss function for the basic neural network model can be pre-set. This loss function can be a cross entropy loss function, a mean square error loss function, an exponential loss function, etc., which are not specifically limited here.

[0124] In some embodiments, the basic neural network model outputs a prediction information embedding vector for the target node. On this basis, the embedding vector of the information indicated by the target node in the reference interaction information sequence can be obtained. Therefore, the function value of the loss function can be calculated based on the embedding vector of the prediction information corresponding to the target node and the embedding vector of the information indicated by the target node in the reference interaction information sequence. The function value of the loss function calculated is the prediction error.

[0125] Step 760: Adjust the parameters of the basic neural network model according to the prediction error and continue training until the pre-training end condition is met.

[0126] According to the calculated prediction error, the parameters of the basic neural network model are adjusted in reverse, and then the process of 730-750 is repeated through the basic neural network model with adjusted parameters, so that the recalculated prediction error makes the loss function of the basic neural network model converge, and then the basic neural network model is repeatedly trained according to the above steps 710-750 using the next reference mutual information sequence until the pre-training end condition is reached.

[0127] In an embodiment, a certain proportion of target nodes in a reference interaction information sequence are occluded to obtain a sample interaction information sequence. The basic neural network model predicts the information indicated by the target node in the reference interaction information sequence based on the context nodes of the occluded nodes in the sample interaction information sequence, thereby realizing self-supervised training of the basic neural network model.

[0128] Figure 8 is a flow chart of an information processing method according to an embodiment of the present application. Figure 8 As shown, the method includes: step 810, obtaining a candidate information set in the target domain. Step 820, predicting the target user's behavior label for each information in the candidate information set using an information processing model; the information processing model is trained according to any embodiment of the above-mentioned information processing model training method.

[0129] In some embodiments, the target user may be a new user in the target domain and also a user in the source domain. Since the information processing model has already learned the user characteristics of the source domain user during training, the user characteristics of the user in the source domain can be directly applied to the target domain as the user characteristics of the target user in the target domain. Based on this, the information processing model uses the user characteristics of the target user in the source domain to predict the target user's behavior label for each piece of information in the candidate information set.

[0130] The behavior label is used to indicate the user behavior triggered by the target user in response to the corresponding information. In some embodiments, the behavior label is used to indicate whether the user will click on the corresponding information. Accordingly, the behavior label includes a label indicating that a click behavior is triggered and a label indicating that a click behavior will not be triggered.

[0131] Step 830: Determine the target information to be sent to the target user from the candidate information set based on the behavior tag.

[0132] Based on the predicted behavior labels corresponding to each information in the candidate information set, information whose behavior labels indicate that the target user will click can be filtered out from the candidate information set, and the filtered information can be used as the target information to be sent to the target user.

[0133] In this solution, the information processing model is trained by the first user behavior data and the second user behavior data in the source domain to ensure that the information processing model can accurately learn the characteristics of the users in the source domain. On this basis, the characteristics of the user learned by the information processing model in the source domain are applied to the target domain, and the characteristics of the user learned in the source domain are used to reflect the user's preference for information. Since the target information to be sent in the target domain is determined by combining the characteristics of the user in the source domain, the matching between the target information determined for the user and the target user is guaranteed. Therefore, it can avoid the situation of multiple information screening and information sending, and effectively solve the problem of low information sending efficiency due to low matching between information and users in the existing technology. Moreover, the solution of this application can also be used to determine target information to users in the cold start scenario of the target domain.

[0134] In some embodiments of the present application, the information processing model includes a trained first neural network model, a trained user feature model, an embedding lookup layer, and a classification layer; Figure 9 As shown, step 820 includes:

[0135] Step 910: Obtain global features of the target user from the user global feature set; the global features in the user global feature set are generated by the trained first neural network model and the trained user feature model according to the first user behavior data.

[0136] As described above, using the trained first neural network model and the trained user feature model, global features for each user in the source domain can be generated based on the first user behavior data in the source domain, and the global features for each user in the source domain can be added to the user global feature set. After the information processing model is applied to the target domain, when information recommendations are needed for a target user, if the target user is both a new user in the target domain and a user in the source domain, the global features of the target user can be retrieved from the user global feature set.

[0137] In some embodiments of the present application, when the source domain and the target domain are divided according to applications, assuming that the source domain corresponds to a first application and the target domain corresponds to a second application, the first application and the second application may be applications with mutually open login accounts. For example, the login account in the first application can be authorized to log in to the second application. In this case, when the account authorized to log in to the second application is the login account in the first application, the user can be considered as both a user in the source domain and a user in the target domain. Thus, the user's login account is used as the user identifier, and the user's global features are searched in the user's global feature set to obtain the user's global features.

[0138] Step 920: The embedding lookup layer generates an embedding vector for each information in the candidate information set. In some embodiments, the embedding lookup layer may generate an embedding vector for each information in the candidate information set using an embedding matrix corresponding to the embedding layer in the first neural network model.

[0139] In step 930 , the classification layer outputs a behavior label of the target user for each piece of information in the candidate information set based on the global features of the target user and the embedding vector of each piece of information in the candidate information set.

[0140] In some embodiments of the present application, the classification layer can output the target user's behavior label for each information in the candidate information set based on the global features of the target user and the embedding vector of each information in the candidate information set through a softmax function.

[0141] Figure 10 FIG. 1 is a schematic diagram of an information processing model predicting behavior labels according to an embodiment of the present application. Figure 10 As shown, the first neural network model 310 outputs the first user features of each user based on the interaction information sequence of each user in the source domain, and then the user feature model 320 performs deep feature extraction from the first user features of each user to obtain the global features of each user in the source domain, and adds the global features of each user in the source domain to the user global feature set.

[0142] On this basis, when it is necessary to determine the target domain information for the target user, the target user's global features are obtained from the user global feature set based on the target user's user ID. The embedding lookup layer 1010 in the information processing model generates an embedding vector for each information in the target domain based on the target domain ID corresponding to each information in the target domain (the target domain ID is used to identify the information in the target domain). Finally, the classification layer 1020 in the information processing model classifies the information based on the target user's global features and the embedding vector corresponding to the information in the target domain, outputting the target user's behavior label for the information in the target domain.

[0143] In some embodiments of the present application, before step 820, the method further includes: receiving an information request, where the information request indicates a user identifier of the target user; in this embodiment, after step 820, the information processing method further includes: sending target information to the target user based on the user identifier of the target user.

[0144] Figure 11 FIG. 1 is a schematic diagram showing target information displayed in a user interface according to an embodiment of the present application. Figure 11 In the corresponding embodiment, the target domain is the domain to which the reading application belongs, wherein the "Take a Look" subpage in the reading application displays the information sent by the reading application server to the user, specifically, the information sent is an article. Specifically, the "Take a Look" subpage can display the title of the sent article (such as Figure 11 Article Title I, Article Title II, and Article Title III), the name of the author of the article (such as Figure 11 Author I, Author II and Author III in the article), further, the cover image of the sent article can also be displayed.

[0145] In a specific embodiment, a rigorous experiment was conducted on the first dataset ColdRec-2 and the second dataset for the information processing model of this solution. In the experiment, the information processing model trained according to the training method of this application was trained. Figure 10 The information processing model shown is compared with other recommendation models. The reference models include NeuFM, DeepFM (a recommendation model proposed by Huawei's Noah's Ark Lab in 2017), MTL, and PeterRec (Parameter-Efficient Transfer from Sequential Behaviors for User Modeling and Recommendation). In the experiment, the performance of each recommendation model was tested using HR@5 as a parameter metric. HR@5, short for Hit Ratio Top 5, refers to the accuracy of the top five items in the recall. The experimental results are shown in Table 1 below.

[0146] Table 1

[0147]

[0148] As shown in Table 1 above, the ETM model is trained according to the training method of this application. Figure 10As shown in Table 1, compared with other models, the information processing model trained by the present application achieved an accuracy improvement of more than 15% on both the first dataset ColdRec-2 and the second dataset, indicating that the information processing model obtained by the training method of the present application effectively improved the accuracy of information determination in the target domain.

[0149] In some embodiments of the present application, after the step of pushing target information to the target user according to the user identification of the target user, the information recommendation method further includes: obtaining the target user's interactive behavior information with respect to the target information; and updating and training the information processing model according to the target user's interactive behavior information with respect to the target information.

[0150] The target user's interactive behavior information for the target information is used to indicate the interactive behavior triggered by the target user for the target information, such as click behavior. The server can send information in the target domain to multiple target users. When the amount of data collected on the target user's interactive behavior information for the target information is sufficient, the information processing model is updated and trained based on the target user's interactive behavior information for the target information. Thus, the information processing model can generate user features for each user in the target domain. Furthermore, in the subsequent process, the information processing model can directly send targeted information based on the user's global features in the target domain, without using the user's global features in the source domain to send information in the target domain.

[0151] The following describes an embodiment of the device of the present application, which can be used to perform the method described in the above embodiment of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the above method embodiment of the present application.

[0152] Figure 12 is a block diagram of a training device for an information processing model according to an embodiment. Figure 12As shown, the training device of the information processing model includes: a first user behavior data acquisition module 1210, which is used to obtain the interaction information sequence of the first user and the interaction information sequence of the second user from the first user behavior data of the source domain; a first output module 1220, which is used to output the first user feature of the first user according to the interaction information sequence of the first user by the first neural network model, and a second output module 1230, which is used to output the first user feature of the second user according to the interaction information sequence of the second user by the first neural network model; the first neural network model is pre-trained and determined based on the historical behavior data of the second user in the source domain; a first generation module 1240, which is used to generate the local feature of the first user and the global feature of the first user according to the first user feature of the first user by the user feature model; and a second generation module 1250. 250, for generating local features of the second user according to the first user features of the second user by the user feature model; a first score calculation module 1260, for calculating a first score according to the local features of the first user and the global features of the first user by the discriminant model; and a second score calculation module 1270, for calculating a second score according to the global features of the first user and the local features of the second user by the discriminant model; a first training module 1280, for training the first neural network model, the user feature model and the discriminant model according to the first score and the second score until the training end condition is met; an information processing model determination module 1290, for determining an information processing model according to the trained first neural network model and the trained user feature model, the information processing model determining the target information to be sent to the user in the target domain.

[0153] In some embodiments of the present application, the user feature model includes at least two convolution layers and a pooling layer, and the length of the convolution window corresponding to each convolution layer in the at least two convolution layers is different; in this embodiment, the first generation module 1240 includes: a first convolution processing unit, which is used to perform convolution processing on the first user feature of the first user according to the corresponding convolution window by each convolution layer in the at least two convolution layers, so as to obtain multiple intermediate local features of the first user; a first splicing unit, which is used to splice the multiple intermediate local features of the first user to obtain the local features of the first user; a first pooling processing unit, which is used to perform pooling processing on the local features of the first user by the pooling layer, so as to obtain the global features of the first user; in this embodiment, the second generation module 1250 includes: a second convolution processing unit, which is used to perform convolution processing on the first user feature of the second user according to the corresponding convolution window by each convolution layer in the at least two convolution layers, so as to obtain multiple intermediate local features of the second user; a second splicing unit, which is used to splice the multiple intermediate local features of the second user to obtain the local features of the second user.

[0154] In some embodiments of the present application, the first neural network model includes a basic neural network model and an adaptive network inserted into the basic neural network model, and the parameters of the basic neural network model are pre-trained and determined based on the historical behavior data of the second user in the source domain; the training device of the information processing model also includes: a parameter fixing module for fixing the parameters of the basic neural network model; in this embodiment, the first training module 1280 is further configured to: adjust the parameters of the adaptive network, the user feature model and the discriminant model according to the first score and the second score and continue training until the training end condition is reached.

[0155] In some embodiments of the present application, the basic neural network model includes an embedding layer and multiple cascaded converter neural networks, and the first neural network model is divided into N cascaded sub-neural networks, each of which includes a converter neural network and at least one adaptive network inserted into the converter neural network; N is an integer greater than 1; the first output module 1220 includes: an input information acquisition unit, configured to acquire input information corresponding to the k-level sub-neural network; wherein, if k=1, the input information corresponding to the k-level sub-neural network is the embedding vector output by the embedding layer according to the interaction information sequence of the first user; if k>1, the input information corresponding to the k-level sub-neural network is the k-1-level intermediate user features of the first user output by the k-1-level sub-neural network; 0<k≤N, k is an integer; the first input unit, configured to input the input information corresponding to the k-level sub-neural network into the k-level sub-neural network; the feature extraction unit, configured to perform feature extraction by the k-level sub-neural network based on the corresponding input information and output the k-level intermediate user features of the first user; wherein, if k=N, the k-level intermediate user features of the first user are used as the first user features of the first user; and if k<N, the k-level intermediate user features of the first user are used as the input information of the k+1-level sub-neural network.

[0156] In some embodiments of the present application, the converter neural network includes a cascaded multi-head attention layer, a first feedforward neural network layer, a first summation and normalization layer, a second feedforward neural network layer, and a second summation and normalization layer; each sub-neural network includes an adaptive network inserted between the first feedforward neural network layer and the first summation and normalization layer, and an adaptive network inserted between the second feedforward neural network layer and the second summation and normalization layer; the sub-neural network also includes an identity mapping from the input of the multi-head attention layer to the input of the first summation and normalization layer, and an identity mapping from the input of the second feedforward neural network layer to the input of the second summation and normalization layer.

[0157] In some embodiments of the present application, the adaptive network includes a cascaded first fully connected layer, an intermediate fully connected layer, an activation layer and a last fully connected layer, the number of neurons in the first fully connected layer is equal to the number of neurons in the last fully connected layer, and the number of neurons in the first fully connected layer is greater than the number of neurons in the intermediate fully connected layer; the adaptive network also includes an identity mapping from the input of the first fully connected layer to the output of the last fully connected layer.

[0158] In some embodiments of the present application, the training device of the information processing model also includes: a target node selection module, which is used to select multiple target nodes in the reference interaction information sequence; the reference interaction information sequence refers to the interaction information sequence of each user in the second user's historical behavior data; a replacement module, which is used to block multiple target nodes in the reference interaction information sequence to obtain a sample interaction information sequence; a sample interaction information sequence input module, which is used to input the sample interaction information sequence into the basic neural network model; a prediction information output module, which is used for the basic neural network model to output prediction information corresponding to the target node according to the sample interaction information sequence; a prediction error determination module, which is used to determine the prediction error based on the prediction information and the target node; a second training module, which is used to adjust the parameters of the basic neural network model according to the prediction error and continue training until the pre-training end condition is reached.

[0159] Figure 13 is a block diagram of an information processing device according to an embodiment of the present application. Figure 13 As shown, the information processing device includes: a candidate information set acquisition module 1310, used to obtain a candidate information set in a target domain; a behavior label prediction module 1320, used to predict the behavior label of the target user for each information in the candidate information set through an information processing model; the information processing model is trained according to the training method of the information processing model in any of the above embodiments; a target information determination module 1330, used to determine the target information to be sent to the target user in the candidate information set based on the behavior label.

[0160] In some embodiments of the present application, the information processing model includes a trained first neural network model, a trained user feature model, an embedding search layer and a classification layer; the behavior label prediction module 1320 includes: a global feature search unit, used to obtain the global features of the target user in the user global feature set; the global features in the user global feature set are generated by the trained first neural network model and the trained user feature model according to the first user behavior data; an embedding vector generation unit, used to generate an embedding vector for each information in the candidate information set by the embedding search layer; a behavior label output unit, used to output the target user's behavior label for each information in the candidate information set by the classification layer based on the global features of the target user and the embedding vector of each information in the candidate information set.

[0161] In some embodiments of the present application, the information processing device further includes: a receiving module for receiving an information request indicating a user identifier of a target user; and a sending module for sending target information to the target user according to the user identifier of the target user.

[0162] In some embodiments of the present application, the information recommendation device also includes: an interactive behavior information acquisition module for acquiring the interactive behavior information of the target user for the target information; and an update training module for updating and training the information processing model based on the interactive behavior information of the target user for the target information.

[0163] Figure 14 The following is a schematic diagram showing the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application. Figure 14 The computer system 1400 of the electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present application.

[0164] like Figure 14 As shown, computer system 1400 includes a central processing unit (CPU) 1401, which can perform various appropriate actions and processes according to programs stored in read-only memory (ROM) 1402 or programs loaded from storage unit 1408 into random access memory (RAM) 1403, such as executing the methods in the above embodiments. Various programs and data required for system operation are also stored in RAM 1403. CPU 1401, ROM 1402, and RAM 1403 are connected to each other via bus 1404. Input / output (I / O) interface 1405 is also connected to bus 1404.

[0165] The following components are connected to the I / O interface 1405: an input section 1406 including a keyboard, a mouse, and the like; an output section 1407 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 1408 including a hard disk; and a communication section 1409 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 1409 performs communication processing via a network such as the Internet. A drive 1410 is also connected to the I / O interface 1405 as needed. Removable media 1411, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 1410 as needed, so that computer programs read from the removable media can be installed in the storage section 1408 as needed.

[0166] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1409, and / or installed from a removable medium 1411. When the computer program is executed by the central processing unit (CPU) 1401, the various functions defined in the system of the present application are executed.

[0167] It should be noted that the computer-readable medium shown in the embodiments of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a 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 above. In the present application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0168] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. Among them, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0169] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. In some cases, the names of these units do not constitute limitations on the units themselves.

[0170] As another aspect, the present application further provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments, or may exist independently without being incorporated into the electronic device. The computer-readable storage medium carries computer-readable instructions, and when the computer-readable instructions are executed by a processor, the method of any of the above embodiments is implemented.

[0171] According to one aspect of the present application, an electronic device is also provided, which includes: a processor; a memory, wherein computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the processor, the method in any of the above embodiments is implemented.

[0172] According to one aspect of an embodiment of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method of any of the above embodiments.

[0173] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the application, the features and functions of two or more modules or units described above can be concretized in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0174] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the embodiments of the present application.

[0175] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of this application and include common knowledge or customary techniques in the art that are not disclosed herein.

[0176] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A training method for an information processing model, characterized in that: include: Obtaining an interaction information sequence of the first user and an interaction information sequence of the second user from the first user behavior data in the source domain; The first neural network model outputs a first user feature of the first user based on the interaction information sequence of the first user, and the first neural network model outputs a first user feature of the second user based on the interaction information sequence of the second user; the first neural network model is pre-trained based on historical behavior data of the second user in a source domain; generating, by a user feature model, a local feature of the first user and a global feature of the first user based on the first user feature of the first user; and generating, by the user feature model, a local feature of the second user based on the first user feature of the second user; Calculating a first score by a discriminant model based on the local features of the first user and the global features of the first user; and calculating a second score by the discriminant model based on the global features of the first user and the local features of the second user; Training the first neural network model, the user feature model, and the discriminant model according to the first score and the second score until a training end condition is met; An information processing model is determined based on the trained first neural network model and the trained user feature model, and the information processing model is used to determine target information to be sent to the user in the target domain.

2. The method according to claim 1, characterized in that The user feature model includes at least two convolution layers and a pooling layer, and the length of the convolution window corresponding to each convolution layer in the at least two convolution layers is different; The generating, by the user feature model, the local feature of the first user and the global feature of the first user according to the first user feature of the first user includes: Each of the at least two convolutional layers performs convolution processing on the first user feature of the first user according to the corresponding convolution window to obtain a plurality of intermediate local features of the first user; splicing the multiple intermediate local features of the first user to obtain a local feature of the first user; The pooling layer performs pooling processing on the local features of the first user to obtain the global features of the first user; Generating the local features of the second user according to the first user features of the second user by the user feature model includes: Each of the at least two convolutional layers performs convolution processing on the first user feature of the second user according to the corresponding convolution window to obtain a plurality of intermediate local features of the second user; Multiple intermediate local features of the second user are concatenated to obtain a local feature of the second user.

3. The method according to claim 1, characterized in that The first neural network model includes a basic neural network model and an adaptive network inserted into the basic neural network model, wherein parameters of the basic neural network model are determined by pre-training based on historical behavior data of a second user in the source domain; Before the first neural network model outputs the first user feature of the first user according to the interaction information sequence of the first user, the method further includes: Fixing the parameters of the basic neural network model; The step of training the first neural network model, the user feature model, and the discriminant model according to the first score and the second score until a training end condition is met includes: According to the first score and the second score, the parameters of the adaptive network, the user feature model and the discriminant model are adjusted and training is continued until a training end condition is met.

4. The method according to claim 3, characterized in that The basic neural network model includes an embedding layer and a plurality of cascaded transformer neural networks, the first neural network model is divided into N cascaded sub-neural networks, the sub-neural networks include a transformer neural network and at least one adaptive network inserted into the transformer neural network; N is an integer greater than 1; Outputting the first user feature of the first user according to the interaction information sequence of the first user by the first neural network model includes: Obtain input information corresponding to a k-level sub-neural network; wherein, if k=1, the input information corresponding to the k-level sub-neural network is the embedding vector output by the embedding layer according to the interaction information sequence of the first user; if k>1, the input information corresponding to the k-level sub-neural network is the k-1-level intermediate user features of the first user output by the k-1-level sub-neural network; 0<k≤N, k is an integer; Inputting the input information corresponding to the k-level sub-neural network into the k-level sub-neural network; The k-level sub-neural network performs feature extraction based on the corresponding input information and outputs the k-level intermediate user features of the first user; wherein, if k=N, the k-level intermediate user features of the first user are used as the first user features of the first user; if k<N, the k-level intermediate user features of the first user are used as the input information of the k+1-level sub-neural network.

5. The method according to claim 4, characterized in that The converter neural network includes a cascaded multi-head attention layer, a first feedforward neural network layer, a first summation and normalization layer, a second feedforward neural network layer, and a second summation and normalization layer; Each of the sub-neural networks includes an adaptive network inserted between the first feedforward neural network layer and the first summation and normalization layer, and an adaptive network inserted between the second feedforward neural network layer and the second summation and normalization layer; The sub-neural network also includes an identity mapping from the input of the multi-head attention layer to the input of the first summation and normalization layer, and an identity mapping from the input of the second feedforward neural network layer to the input of the second summation and normalization layer.

6. The method according to claim 5, characterized in that The adaptive network includes a cascaded first fully connected layer, an intermediate fully connected layer, an activation layer and a last fully connected layer, the number of neurons in the first fully connected layer is equal to the number of neurons in the last fully connected layer, and the number of neurons in the first fully connected layer is greater than the number of neurons in the intermediate fully connected layer; the adaptive network also includes an identity mapping from the input of the first fully connected layer to the output of the last fully connected layer.

7. The method according to any one of claims 3 to 6, characterized in that Before the first neural network model outputs the first user feature of the first user according to the interaction information sequence of the first user, the method further includes: Selecting multiple target nodes in a reference interaction information sequence; the reference interaction information sequence refers to an interaction information sequence of each user in the second user historical behavior data; Obstructing the plurality of target nodes in the reference interaction information sequence to obtain a sample interaction information sequence; Inputting the sample interaction information sequence into the basic neural network model; The basic neural network model outputs prediction information corresponding to the target node according to the sample interaction information sequence; determining a prediction error based on the prediction information and the target node; The parameters of the basic neural network model are adjusted according to the prediction error and training is continued until the pre-training end condition is reached.

8. An information processing method, characterized in that: include: Obtain a candidate information set in the target domain; Predicting the target user's behavior label for each piece of information in the candidate information set by using the information processing model; The information processing model is trained according to the method according to any one of claims 1 to 7; According to the behavior tag, target information to be sent to the target user is determined in the candidate information set.

9. The method according to claim 8, characterized in that The information processing model includes a trained first neural network model, a trained user feature model, an embedding search layer, and a classification layer; and predicting the target user's behavior label for each piece of information in the candidate information set by the information processing model includes: Obtaining global features of the target user from a user global feature set; the global features in the user global feature set are generated by the trained first neural network model and the trained user feature model according to the first user behavior data; The embedding lookup layer generates an embedding vector for each information in the candidate information set; The classification layer outputs a behavior label of the target user for each piece of information in the candidate information set based on the global features of the target user and the embedding vector of each piece of information in the candidate information set.

10. The method according to claim 8, characterized in that Before predicting the target user's behavior label for each piece of information in the candidate information set by using the information processing model, the method further includes: receiving an information request, wherein the information request indicates a user identifier of the target user; After determining the target information to be sent to the target user from the candidate information set based on the behavior tag, the method further includes: The target information is sent to the target user according to the user identifier of the target user.

11. The method according to claim 10, characterized in that After sending the target information to the target user according to the user identifier of the target user, the method further includes: Obtaining interactive behavior information of the target user with respect to the target information; The information processing model is updated and trained according to the target user's interactive behavior information with respect to the target information.

12. A training device for an information processing model, characterized in that: include: A first user behavior data acquisition module is configured to acquire an interaction information sequence of a first user and an interaction information sequence of a second user from the first user behavior data in a source domain; a first output module, configured to output, by a first neural network model, a first user feature of the first user based on the interaction information sequence of the first user; and a second output module, configured to output, by the first neural network model, a first user feature of the second user based on the interaction information sequence of the second user; the first neural network model being pre-trained and determined based on historical behavior data of the second user in a source domain; a first generating module configured to generate, by a user feature model, a local feature of the first user and a global feature of the first user based on the first user feature of the first user; and a second generating module configured to generate, by the user feature model, a local feature of the second user based on the first user feature of the second user; A first score calculation module, configured to calculate a first score according to the local features of the first user and the global features of the first user by a discriminant model; and a second score calculation module, configured to calculate a second score by the discriminant model according to the global features of the first user and the local features of the second user; a first training module, configured to train the first neural network model, the user feature model, and the discriminant model according to the first score and the second score until a training end condition is met; An information processing model determination module is used to determine an information processing model based on the trained first neural network model and the trained user feature model, wherein the information processing model determines target information to be sent to the user in the target domain.

13. An information processing device, characterized in that: include: A candidate information set acquisition module is used to acquire a candidate information set in the target domain; a behavior label prediction module, configured to predict the target user's behavior label for each piece of information in the candidate information set using the information processing model; the information processing model is trained according to the method of any one of claims 1 to 7; A target information determination module is used to determine the target information to be sent to the target user in the candidate information set according to the behavior tag.

14. An electronic device, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 11 is implemented.

15. A computer-readable storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, the method according to any one of claims 1 to 11 is implemented.

16. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the method according to any one of claims 1 to 11 is implemented.

Images