Information selection method, device, electronic device and storage medium

Abstract

The present application discloses an information selection method, device, electronic device and storage medium. The embodiments of the present application relate to technologies such as machine learning of artificial intelligence, and the embodiments of the present invention can be applied to various scenarios such as cloud technology, artificial intelligence, smart travel, and assisted driving. The method includes: obtaining a target selection model corresponding to the information category of each target information; inputting the digitized vector into each target selection model to obtain a first probability; inputting the digitized vector and the intervention factor into the target selection model to obtain a second probability; obtaining the gain of each target information according to the first probability and the second probability corresponding to each target information; obtaining the target information with the highest gain as the final selected information. Through the method of the present application, the gain obtained according to the first probability and the second probability can accurately reflect the target user's demand for the target information, thereby making the final selected information more accurate and improving the utilization rate of the information.

Description

Technical Field

[0001] The present application relates to the field of Internet information processing technology, and more specifically, to an information selection method, device, electronic device and storage medium. Background Art

[0002] In the information recommendation service, push information that meets the user's needs can be selected from multiple candidate information and pushed to the user so that the user can directly browse the push information. Candidate information can include news, novels, products, and multimedia information.

[0003] Currently, the user's browsing history and personal information and other features can be input into the trained information selection model, which will output the probability of each candidate information being selected by the user, and select the information with the highest probability as the push information to the user.

[0004] However, since users' browsing behavior cannot accurately reflect their demand for information, the efficiency of information utilization is low and there is a large waste of resources. Summary of the Invention

[0005] In view of this, embodiments of the present application provide an information selection method, device, electronic device, and storage medium.

[0006] In a first aspect, an embodiment of the present application provides an information selection method, the method comprising: obtaining user characteristics of a target user, multiple target information, and a target selection model corresponding to an information category of each target information; inputting the user characteristics of the target user into each target selection model to obtain a first probability corresponding to each target selection model, the first probability being used to characterize the probability that the target information of the information category corresponding to each target selection model is selected by the target user; inputting the user characteristics of the target user and an intervention factor into the target selection model to obtain a second probability corresponding to each target selection model, the second probability being used to characterize the probability that the target information is selected by the target user when intervention is applied to the target information of the information category corresponding to each target selection model, the intervention being used to increase the exposure of the target information; obtaining the first probability and the second probability corresponding to each target information, and obtaining the gain of each target information based on the first probability and the second probability corresponding to each target information; obtaining the target information with the highest gain among the multiple target information as the final selected information for the corresponding target user.

[0007] In a second aspect, an embodiment of the present application provides an information selection device, which includes: a first acquisition module for acquiring user characteristics of a target user, multiple target information, and a target selection model corresponding to the information category of each target information; a first acquisition module for inputting the user characteristics of the target user into each target selection model to obtain a first probability corresponding to each target selection model, and the first probability is used to characterize the probability that the target information of the information category corresponding to each target selection model is selected by the target user; a second acquisition module for inputting the user characteristics of the target user and an intervention factor into the target selection model to obtain a second probability corresponding to each target selection model, and the second probability is used to characterize the probability that the target information is selected by the target user when intervention is applied to the target information of the information category corresponding to each target selection model, and the intervention is used to increase the exposure of the target information; a second acquisition module for acquiring the first probability and the second probability corresponding to each target information, and obtaining the gain of each target information according to the first probability and the second probability corresponding to each target information; an information acquisition module for acquiring the target information with the highest gain among the multiple target information as the final selected information corresponding to the target user.

[0008] Optionally, the second acquisition module is also used to calculate the difference between the second probability corresponding to each target information and the first probability as the probability difference of each target information; obtain the estimated click-through rate corresponding to each target information; calculate the product of the probability difference corresponding to each target information and the estimated click-through rate as the gain of each target information.

[0009] Optionally, the second acquisition module is also used to process the user characteristics of the target user through each target selection model to obtain the user feature vector corresponding to each target selection model; process the intervention factor through each target selection model to obtain the intervention feature vector corresponding to each target selection model; perform attention mechanism operation on the user feature vector and intervention feature vector corresponding to each target selection model to obtain the attention operation result corresponding to each target selection model; process the corresponding user feature vector and attention operation result through each target selection model to obtain the predicted probability output by each target selection model as the second probability corresponding to each target selection model.

[0010] Optionally, the second acquisition module is also used to obtain the key vector and value vector corresponding to each target selection model based on the user feature vector corresponding to each target selection model; calculate the product of the key vector and the intervention feature vector corresponding to each target selection model as the product result corresponding to each target selection model; regularize the product result corresponding to each target selection model to obtain the regularized result corresponding to each target selection model; mask the regularized result corresponding to each target selection model to obtain the mask processing result corresponding to each target selection model; normalize the mask processing result corresponding to each target selection model to obtain the normalized result corresponding to each target selection model; calculate the product of the normalized result corresponding to each target selection model and the value vector as the attention operation result corresponding to each target selection model.

[0011] Optionally, the second acquisition module is also used to splice the vectors of multiple dimensions included in the user feature vector corresponding to each target selection model to obtain the splicing vector corresponding to each target selection model; calculate the sum of the splicing vector corresponding to each target selection model and the attention operation result as the sum vector corresponding to each target selection model; and process the corresponding sum vector through each target selection model to obtain the predicted probability output by each target selection model.

[0012] Optionally, the second acquisition module is also used to normalize the sum vector corresponding to each target selection model to obtain the normalized vector corresponding to each target selection model; through each target selection model, the corresponding normalized vector is processed to obtain the predicted probability output by each target selection model.

[0013] Optionally, the second acquisition module is also used to process the corresponding normalized vectors through each target selection model to obtain the first vector corresponding to each target selection model; concatenate the first vector and the intervention vector corresponding to each target selection model to obtain the second vector corresponding to each target selection model; and process the corresponding second vector through each target selection model to obtain the predicted probability output by each target selection model.

[0014] Optionally, the device also includes a display module; the first acquisition module is also used to obtain the user characteristics of the target user, multiple target information and the target selection model corresponding to the information category of each target information in response to the target user's access request for the target page; the display module is used to display the final selected information on the target page.

[0015] Optionally, the first acquisition module is further used to determine the target selection model corresponding to each target information from multiple preset selection models according to the information category of each target information, and the multiple preset selection models correspond to different information categories respectively.

[0016] Optionally, the device also includes a training module for obtaining training samples corresponding to each information category. The training samples corresponding to each information category include user characteristics of the first user, user characteristics of the second user, a first selection result corresponding to the first sample information under the information category, and a second selection result corresponding to the second sample information under the information category. The first selection result refers to the selection result of the first user for the first sample information when no intervention is applied to the first sample information, and the second selection result refers to the selection result of the second user for the second sample information when intervention is applied to the second sample information. The initial model is trained according to the training samples corresponding to each information category to obtain a target selection model corresponding to each information category.

[0017] In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor and a memory; one or more programs are stored in the memory and configured to be executed by the processor to implement the above method.

[0018] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which program code is stored, wherein the above method is executed when the program code is run by a processor.

[0019] In a fifth aspect, embodiments of the present application provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the above-described method.

[0020] An information selection method, device, electronic device and storage medium provided in the embodiments of the present application obtain a first probability and a second probability corresponding to each target information through a target selection model, wherein the second probability reflects the probability that the target information is selected by the target user after intervention is applied to the target information. The gain obtained based on the first probability and the second probability can accurately reflect the target user's demand for the target information, thereby making the final selected information determined based on the gain highly matched with the user's needs, thereby improving information utilization efficiency and reducing resource waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 is a schematic diagram of an application scenario according to an embodiment of the present application;

[0023] Figure 2 A flowchart of an information selection method proposed in one embodiment of the present application is shown;

[0024] Figure 3 A flowchart of an information selection method proposed in another embodiment of the present application is shown;

[0025] Figure 4 A schematic diagram of the structure of a target selection model in an embodiment of the present application is shown;

[0026] Figure 5 A schematic diagram of the structure of another target selection model in an embodiment of the present application is shown;

[0027] Figure 6 A flowchart of an information selection method proposed in another embodiment of the present application is shown;

[0028] Figure 7 A schematic diagram of the structure of another target selection model in an embodiment of the present application is shown;

[0029] Figure 8 A flowchart of an information selection method proposed in another embodiment of the present application is shown;

[0030] Figure 9 A schematic diagram of a target page in an embodiment of the present application is shown;

[0031] Figure 10 A flowchart of an information selection method proposed in another embodiment of the present application is shown;

[0032] Figure 11 A schematic diagram of the training process of the preset selection model in an embodiment of the present application is shown;

[0033] Figure 12 A schematic diagram showing the information selection process in an embodiment of the present application is shown;

[0034] Figure 13 A schematic diagram of the structure of another target selection model in an embodiment of the present application is shown;

[0035] Figure 14A block diagram of an information selection device proposed in one embodiment of the present application is shown;

[0036] Figure 15 A structural block diagram of an electronic device for executing the information selection method according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0037] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of this application without making creative efforts are within the scope of protection of this application.

[0038] In the following description, the terms "first\second" are only used to distinguish similar objects and do not represent a specific order for the objects. It can be understood that "first\second" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[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] The present application discloses an information selection method, device, electronic device and storage medium, which involve artificial intelligence machine learning and other technologies.

[0042] 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 involves studying the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0043] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0044] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0045] Reinforcement Learning (RL), also known as reinforcement learning, evaluation learning, or enhanced learning, is a paradigm and methodology in machine learning. It is used to describe and solve the problem of how an agent learns strategies to maximize rewards or achieve specific goals during its interaction with the environment.

[0046] A common model for reinforcement learning is the standard Markov decision process (MDP). Depending on the given conditions, reinforcement learning can be divided into model-based reinforcement learning (RL) and model-free reinforcement learning (RL), as well as active reinforcement learning (RL) and passive reinforcement learning (RL). Reinforcement learning variants include inverse reinforcement learning, hierarchical reinforcement learning, and reinforcement learning for partially observable systems. Algorithms used to solve reinforcement learning problems can be categorized into policy search algorithms and value function algorithms. Deep learning models can be used in reinforcement learning, forming deep reinforcement learning.

[0047] The number of daily active users (DAU) is a type of data used to reflect the operating conditions and user scale of Internet applications, websites, etc.

[0048] Business DAUs specifically refer to the number of daily active users (DAUs) of a specific sub-service within an internet application (e.g., a mobile app) that includes multiple sub-services. For example, a browser includes services such as information feeds, free novels, search, files, and tools. The DAUs of these services collectively constitute the browser DAUs.

[0049] Business penetration rate refers to the ratio of business DAU to application-side DAU, reflecting the penetration of different sub-businesses in the overall users of Internet applications.

[0050] Resource placement refers to the location on a specific page of an internet application where recommended information can be displayed. For example, when a user is browsing images in a browser, recommended novels can be displayed to the user through the resource placement in the browser.

[0051] A randomized experiment involves randomly dividing a population into an intervention group and a control group. An intervention is administered to the intervention group, while no intervention is administered to the control group. The randomized data generated by a randomized experiment can demonstrate the causal effect of the intervention on the population as a whole.

[0052] The causal effect refers to the comparison of the observed results when an intervention occurs and the observed results when the intervention does not occur. If the two observed results are different, the intervention is said to have a causal effect on the results, and the difference between the observed results under intervention and without intervention is the size of the causal effect.

[0053] The Response model directly models the user's response based on user characteristics. For example, based on the user's historical data, it models whether the user will click on the recommended information, where a click is the user's response.

[0054] It should be noted that in the embodiments of the present application, the acquisition of user characteristics such as user browsing history, click history, purchase history, collection history, and age requires the user's permission or consent, and the collection, use, processing and storage of user characteristics such as user browsing history, click history, purchase history, collection history, and age must comply with the regulations of the region.

[0055] like Figure 1 As shown, the application scenario applicable to the embodiment of the present application includes a terminal 20 and a server 10, which are connected to each other via a wired network or a wireless network. The terminal 20 can be a smartphone, tablet computer, laptop computer, desktop computer, smart home appliance, vehicle terminal, aircraft, wearable device terminal, virtual reality device, or other terminal device capable of page display, or running other applications that can call page display applications (such as instant messaging applications, shopping applications, search applications, game applications, forum applications, map and transportation applications, etc.).

[0056] The server 10 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or 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), and big data and artificial intelligence platforms. The server 10 can be used to provide services for applications running on the terminal 20.

[0057] Among them, the terminal 20 can send the user's display requirements to the server 10, so that the server 10 can feedback the corresponding display page according to the user's display requirements, or the server 10 can directly send the display page to the terminal 20, or the server 10 can send a user-customized display page or a display page related to the user's interests based on the user information bound to the terminal 20.

[0058] The terminal 20 can receive an access request from the target user and send the access request to the service 10. The server 10 responds to the access request and, based on the user characteristics and intervention factors of the target user, determines the first probability and the second probability corresponding to each of the multiple target information through the target selection model, and determines the gain of each target information based on the first probability and the second probability of each target information. The final selected information is then screened from the multiple target information, and the server 10 sends the final selected information to the terminal 10 so that the terminal 20 can display the final selected information to the target user.

[0059] The target user can be the user for whom the information is ultimately selected. For example, the target user can be a child, an elderly person, or a young person. The target user can also be a white-collar worker, a doctor, or a programmer. User characteristics of the target user can include information such as the target user's browsing history, click history, purchase history, favorite history, and age.

[0060] The target information may refer to candidate information. The target information may be any physical item, virtual item, or multimedia information that can be selected. For example, the target information may be a physical commodity, a virtual commodity, news, search information, or a novel.

[0061] In another embodiment, the target selection model is preset in the terminal 20. The terminal 20 can receive an access request from the target user and, in response to the access request, determine the first probability and the second probability corresponding to each of the multiple target information based on the user characteristics and intervention factors of the target user through the target selection model, and determine the gain of each target information based on the first probability and the second probability of each target information, and then screen the final selected information from the multiple target information. Finally, the terminal 20 displays the final selected information to the target user.

[0062] For the convenience of description, in the following embodiments, information selection is performed by an electronic device as an example.

[0063] See also Figure 2 , Figure 2 A flowchart of an information selection method proposed in one embodiment of the present application is shown. The method can be applied to an electronic device, which can be Figure 1 The server 10 in the embodiment of the present invention comprises:

[0064] S110: Obtain user characteristics of a target user, multiple target information, and a target selection model corresponding to an information category of each target information.

[0065] In this embodiment, the target information may include novels, news information, and search information. For example, a novel, a news information, and a search information may be used as a target information. The user characteristics of the target user are as described above and will not be repeated here.

[0066] Different target information can correspond to different information categories, including novels, news, and search categories. Each information category has its own target selection model. For example, a novel corresponds to the novel category, and its target selection model is the novel category's target selection model. News information corresponds to the news category, and its target selection model is the news category's target selection model.

[0067] As an implementation method, the target selection model corresponding to each target information can be determined from a plurality of preset selection models according to the information category of each target information, and the plurality of preset selection models correspond to different information categories, respectively.

[0068] The electronic device can store multiple preset selection models, one for each information category. For target information, the electronic device obtains the information category of the target information and uses the preset selection model corresponding to the information category as the target selection model for the target information. Different target information may correspond to different target selection models.

[0069] For example, the preset selection models include a preset selection model a1 corresponding to the search category, a preset selection model a2 corresponding to the novel category, and a preset selection model a3 corresponding to the information category. The target information includes 1 novel b and 1 information information c. Then, the target selection model corresponding to the target information b is determined to be a1, and the target selection model corresponding to the target table information c is determined to be a3.

[0070] S120. Input the user characteristics of the target user into each target selection model to obtain a first probability corresponding to each target selection model. The first probability is used to represent the probability that the target information of the information category corresponding to each target selection model is selected by the target user.

[0071] The user features of the target user can be digitized into a vector, and the vector after the digitization of the user features is input into each target selection model to obtain the predicted probability output by each target selection model as the first probability corresponding to each target selection model.

[0072] For target information within the same information category, the target selection model for that information category determines that the probability of each target information within that information category being selected by the target user is the first probability. For example, the target information corresponding to the novel category is 10 novels. The target selection model for the novel category processes the user features of the target user, and the predicted probability output by the target selection model for the novel category is used as the first probability. The probabilities of the target users selecting these 10 novels are all the first probabilities output by the target selection model for the novel category.

[0073] S130. Input the user characteristics and intervention factors of the target user into the target selection model to obtain the second probability corresponding to each target selection model. The second probability is used to characterize the probability that the target information is selected by the target user when intervention is applied to the target information of the information category corresponding to each target selection model. The intervention is used to increase the exposure of the target information.

[0074] The user features of the target user can be digitized into a vector, and the vector after the digitization of the user features and the intervention factor are input into each target selection model to obtain the predicted probability output by each target selection model as the second probability corresponding to each target selection model.

[0075] For target information within the same information category, the target selection model for that information category determines that the probability of each target information within that information category being selected by the target user is the second probability. For example, if the target information for an information category is five pieces of information, the target selection model for that information category processes the target user's user characteristics and intervention factors, and the predicted probability output by the target selection model for that information category is used as the second probability. The probability of each of the five pieces of information being selected by the target user is the second probability output by the target selection model for that information category.

[0076] Intervention refers to certain operations imposed on the test users (such as issuing coupons and sending marketing emails). In this embodiment, intervention refers to sending target information to target users, thereby increasing the exposure of the target information.

[0077] In this embodiment, the intervention factor may refer to 1. When the user characteristics of the target user and the intervention factor are input into the target selection model, it represents that intervention is applied to the target information corresponding to the target selection model. At this time, the second probability obtained represents the probability that the target information is selected by the target user when intervention is applied to the target information of the corresponding information category of each target selection model.

[0078] In some possible implementations, a digitized vector of user features and a non-interference factor may be input into each target selection model to obtain a first probability output by each target selection model, where the non-interference factor may be 0. When the user features and the non-interference factor of the target user are input into the target selection model, no intervention is applied to the target information corresponding to the target selection model. In this case, the obtained first probability represents the probability that the target information will be selected by the target user when no intervention is applied to the target information of the information category corresponding to each target selection model.

[0079] As an implementation method, a training method for a target selection model (the training method for a preset selection model refers to the training method for a target selection model and will not be repeated here) includes: obtaining training samples corresponding to each information category, the training samples corresponding to each information category include user characteristics of the first user, user characteristics of the second user, a first selection result corresponding to the first sample information under the information category, and a second selection result corresponding to the second sample information under the information category, the first selection result refers to the selection result of the first user for the first sample information when no intervention is applied to the first sample information, and the second selection result refers to the selection result of the second user for the second sample information when intervention is applied to the second sample information; the initial model is trained according to the training samples corresponding to each information category to obtain a target selection model corresponding to each information category.

[0080] For each information category, the training samples corresponding to the information category may include the user characteristics of the first user, the user characteristics of the second user, the first selection results corresponding to each of the multiple first sample information under the information category, and the second selection results corresponding to each of the multiple second sample information under the information category; the user characteristics of the first user, the non-intervention factors, and the first selection results corresponding to each of the multiple first sample information under the information category may be used as non-intervention training samples to train the initial model; at the same time, the user characteristics of the second user, the intervention factors, and the second selection results corresponding to each of the multiple second sample information under the information category may be used as intervention training samples to train the initial model; when both the non-intervention training samples and the intervention training samples have completed training, the initial model is used as the target selection model corresponding to the information category.

[0081] The user characteristics and non-intervention factors of the first user may be input into the initial model to obtain the probability of selection by the first user for each of the plurality of first sample information output by the initial model. A loss value may be determined based on the probability of selection by the first user for each of the plurality of first sample information and the first selection results for each of the plurality of first sample information. The initial model may be trained using the determined loss value, thereby achieving training of the initial model using non-intervention training samples. Simultaneously, the user characteristics and intervention factors of the second user may be input into the initial model to obtain the probability of selection by the second user for each of the plurality of second sample information output by the initial model. A loss value may be determined based on the probability of selection by the second user for each of the plurality of second sample information and the second selection results for each of the plurality of second sample information. The initial model may be trained using the determined loss value, thereby achieving training of the initial model using intervention training samples.

[0082] In this embodiment, the trained target selection model is an Uplift model. An Uplift model is a model derived from modeling causal events based on randomized experiments. The model is trained by analyzing the performance of different groups of subjects under intervention and without intervention. The trained model then predicts changes in users after the intervention (i.e., the causal effect of the intervention). For example, in certain scenarios (such as marketing), the focus is not just on user responses, but rather on the changes in responses after a certain intervention, such as the increase in the probability of a user purchasing after receiving a coupon. This type of problem cannot be solved using a Response model, but the Uplift model can address this problem.

[0083] It can be understood that through the target selection model of the present application, the change in the probability of the target table information being selected by the target user before and after the target information is intervened (the difference between the first probability and the second probability) can be accurately predicted, so that it can be determined whether to push the target information based on the change in the probability.

[0084] S140 , obtaining a first probability and a second probability corresponding to each target information, and obtaining a gain of each target information according to the first probability and the second probability corresponding to each target information.

[0085] The target user's user characteristics are input into each target selection model to obtain a first probability corresponding to each target selection model. The target user's user characteristics and intervention factors are also input into the target selection model to obtain a second probability corresponding to each target selection model. For each target information, the first probability and second probability output by the target selection model corresponding to the information category of the target information are obtained as the first probability and second probability corresponding to the target information.

[0086] As an embodiment, a method for determining the gain of each target information based on the first probability and the second probability corresponding to each target information may include: calculating the difference between the second probability and the first probability corresponding to each target information as the probability difference of each target information; obtaining the estimated click-through rate corresponding to each target information; and calculating the product of the probability difference and the estimated click-through rate corresponding to each target information as the gain of each target information.

[0087] As mentioned above, the calculation process of the gain of each target information is expressed with reference to Formula 1, which is as follows:

[0088] (one)

[0089] in, is the gain of target information i, is the estimated click-through rate of the target information, is the probability difference of target information i.

[0090] The calculation formula for the probability difference of target information i can refer to Formula 2, which is as follows:

[0091] (two)

[0092] in, is the second probability of target information i, is the first probability of the target information. The intervention factor is 1. Indicates that the non-intervention factor is 0.

[0093] The estimated click rate corresponding to each target information can be obtained from the server (for example Figure 1 The server can determine the estimated click-through rate of each target information based on a preset algorithm. For example, the preset algorithm can be to average the historical click-through rates of the target information and use the average click-through rate as the preset click-through rate. The preset algorithm can also be a linear interpolation method that uses a linear interpolation method and historical click-through rates to determine the estimated click-through rate.

[0094] In this embodiment, the gain of each target information is determined based on the first and second probabilities of each target information. Because the second probability accurately reflects the probability of the target information being selected by the user after the intervention is applied to the target information, the gain of each target information can accurately reflect the degree of user demand for the target information. The higher the gain of a target information, the higher the target user's demand for the target information.

[0095] S150: Obtain target information with the highest gain among multiple target information as final selection information corresponding to the target user.

[0096] After obtaining the gains of multiple target information, the target information with the highest gain is selected from each target information as the final selected information. The gain of the final selected information is the highest, and the final selected information is the target information that the target user is most likely to pay attention to.

[0097] As an implementation method, after obtaining the final selected information, the final selected information can be displayed to the target user. After the final selected information is displayed, the probability of the final selected information being selected by the target user is the highest among all target information.

[0098] In this embodiment, the first probability and the second probability corresponding to each target information are obtained through the target selection model, wherein the second probability reflects the probability that the target information is selected by the target user after intervention is applied to the target information. The gain obtained according to the first probability and the second probability can accurately reflect the target user's demand for the target information, thereby making the final selected information determined according to the gain more compatible with the user's needs, thereby improving information utilization efficiency and reducing resource waste.

[0099] At the same time, while improving the accuracy of the final selected information, it can also increase DAU increments and business penetration. Therefore, compared with the existing Response model, the final selected information obtained by the target selection model of this embodiment is more likely to be responded to by users.

[0100] See also Figure 3 , Figure 3 A flowchart of an information selection method proposed in another embodiment of the present application is shown. The method can be applied to an electronic device, which can be Figure 1 The server 10 in the embodiment of the present invention comprises:

[0101] S210: Obtain user characteristics of a target user, multiple target information, and a target selection model corresponding to an information category of each target information.

[0102] S220. Input the user characteristics of the target user into each target selection model to obtain a first probability corresponding to each target selection model. The first probability is used to represent the probability that the target information of the information category corresponding to each target selection model is selected by the target user.

[0103] The description of S210-S220 refers to the description of S110-S120 above and will not be repeated here.

[0104] S230. Process the user features of the target user through each target selection model to obtain a user feature vector corresponding to each target selection model; and process the intervention factor through each target selection model to obtain an intervention feature vector corresponding to each target selection model.

[0105] Each target selection model may include a feature representation module, which is used to convert the user features of the target user and the intervention factor into a vector form to obtain a user feature vector and an intervention feature vector corresponding to each target selection model. The intervention factor may be 1.

[0106] like Figure 4 As shown, the target selection model includes a feature representation module, which includes a user feature unit and a factor unit. The user feature unit and the factor unit in the feature representation module can both be composed of the input layer of the neural network, so as to transform the user features of the target user (including d-dimensional features, d is a non-zero natural number, respectively represented as 、 、…、 ) After inputting the user feature unit, the user feature vector is obtained, and after inputting the intervention factor into the factor unit, the intervention feature vector is obtained.

[0107] S240. Perform an attention mechanism operation on the user feature vector and intervention feature vector corresponding to each target selection model to obtain an attention operation result corresponding to each target selection model.

[0108] An attention mechanism is applied to the user feature vector and intervention feature vector corresponding to each target selection model to obtain an attention operation result. This attention mechanism effectively integrates the user feature vector and intervention feature vector, ensuring that the obtained attention operation result accurately reflects the state of the target information after the intervention, thereby effectively improving the accuracy of the second probability obtained from the attention operation result.

[0109] like Figure 4 As shown, the target selection model also includes an attention mechanism module, which performs attention mechanism operations on the user feature vector and the intervention feature vector to obtain the attention operation results corresponding to the target selection model.

[0110] As an implementation method, S240 may include: obtaining the key vector and value vector corresponding to each target selection model based on the user feature vector corresponding to each target selection model; calculating the product of the key vector and the intervention feature vector corresponding to each target selection model as the product result corresponding to each target selection model; regularizing the product result corresponding to each target selection model to obtain the regularized result corresponding to each target selection model; masking the regularized result corresponding to each target selection model to obtain the masked result corresponding to each target selection model; normalizing the masked result corresponding to each target selection model to obtain the normalized result corresponding to each target selection model; calculating the product of the normalized result corresponding to each target selection model and the value vector as the attention operation result corresponding to each target selection model.

[0111] like Figure 5 As shown, the attention mechanism module of the target selection model may include: a first product unit, used to calculate the product of the key vector and the intervention feature vector corresponding to the target selection model, as the product result corresponding to the target selection model; a regularization unit, used to regularize the product result corresponding to the target selection model to obtain the regularization result corresponding to the target selection model; a masking unit, used to mask the regularization result corresponding to the target selection model to obtain the masking result corresponding to the target selection model; a first normalization unit, used to normalize the masking result corresponding to the target selection model to obtain the normalization result corresponding to the target selection model; a second product unit, used to calculate the product of the normalization result corresponding to the target selection model and the value vector, as the attention operation result corresponding to the target selection model. Among them, the first normalization unit can be masking the regularization result corresponding to the target selection model according to the softmax activation function to obtain the masking result corresponding to the target selection model.

[0112] As mentioned above, the calculation process of the attention operation results corresponding to each target selection model is expressed in Formula 3, which is as follows:

[0113] (three)

[0114] in, is the result of attention operation, is the dimension of the key vector, Q is the intervention feature vector, K is the key vector, and V is the value vector.

[0115] S250. Through each target selection model, the corresponding user feature vector and attention operation result are processed to obtain the predicted probability output by each target selection model as the second probability corresponding to each target selection model.

[0116] After obtaining the corresponding attention operation results and user feature vectors for each target selection model, the attention operation results and user feature vectors corresponding to each target selection model can be summed, and the summed results can be processed by the target selection model to obtain the second probability of the output.

[0117] like Figure 4 Shown and Figure 5 The target selection model also includes an output prediction module, which uses the attention mechanism module to sum the user feature vector corresponding to the target selection model and the attention operation result to obtain the sum result, and then processes the sum result through the output prediction module to obtain the second probability output by the target selection model.

[0118] S260: Obtain a first probability and a second probability corresponding to each target information, and obtain a gain of each target information according to the first probability and the second probability corresponding to each target information.

[0119] S270: Obtain target information with the highest gain among multiple target information as final selection information corresponding to the target user.

[0120] The description of S260-S270 refers to the description of S140-S150 above and will not be repeated here.

[0121] In this embodiment, an attention mechanism operation is performed on the user feature vector and the intervention feature vector corresponding to each target selection model to obtain an attention operation result. Through the attention mechanism operation, the user feature vector and the intervention feature vector can be better fused, so that the obtained attention operation result accurately reflects the corresponding state of the target information after the intervention is applied, thereby effectively improving the accuracy of the second probability obtained according to the attention operation result.

[0122] See also Figure 6 , Figure 6 A flowchart of an information selection method proposed in another embodiment of the present application is shown. The method can be applied to an electronic device, which can be Figure 1 The server 10 in the embodiment of the present invention comprises:

[0123] S310: Obtain user characteristics of a target user, multiple target information, and a target selection model corresponding to an information category of each target information.

[0124] S320. Input the user characteristics of the target user into each target selection model to obtain a first probability corresponding to each target selection model. The first probability is used to represent the probability that the target information of the information category corresponding to each target selection model is selected by the target user.

[0125] S330. Process the user features of the target user through each target selection model to obtain a user feature vector corresponding to each target selection model; and process the intervention factor through each target selection model to obtain an intervention feature vector corresponding to each target selection model.

[0126] S340: Perform an attention mechanism operation on the user feature vector and intervention feature vector corresponding to each target selection model to obtain an attention operation result corresponding to each target selection model.

[0127] Among them, the description of S310-S340 refers to the description of S210-S240 above, and will not be repeated here.

[0128] S350. Splice the vectors of multiple dimensions included in the user feature vector corresponding to each target selection model to obtain a splicing vector corresponding to each target selection model; calculate the sum of the splicing vector corresponding to each target selection model and the attention operation result as the sum vector corresponding to each target selection model.

[0129] Each target selection model processes the input user features in multiple dimensions to obtain a vector corresponding to each target selection model in multiple dimensions. This vector serves as the user feature vector corresponding to that target selection model. Each target selection model concatenates the vectors in multiple dimensions included in the user feature vector to obtain a concatenated vector, which serves as the concatenated vector corresponding to each target selection model.

[0130] After obtaining the splicing vector and attention operation result corresponding to each target selection model, the splicing vector and attention operation result of each target selection model are summed to obtain the sum vector corresponding to each target selection model.

[0131] like Figure 7As shown, the attention mechanism module can also include a first splicing unit and a summation unit. The first splicing unit is used to splice the vectors of multiple dimensions included in the user feature vector corresponding to each target selection model to obtain the splicing vector corresponding to each target selection model; the summation unit is used to calculate the sum of the splicing vector corresponding to each target selection model and the attention operation result as the sum vector corresponding to each target selection model.

[0132] S360. Process the corresponding sum vectors through each target selection model to obtain the predicted probability output by each target selection model as the second probability corresponding to each target selection model.

[0133] After obtaining the sum vector, each target selection model is used to process the corresponding sum vector to obtain the predicted probability output by each target selection model as the second probability corresponding to each target selection model.

[0134] As an implementation method, S360 may include: normalizing the sum vector corresponding to each target selection model to obtain a normalized vector corresponding to each target selection model; processing the corresponding normalized vectors through each target selection model to obtain the predicted probability output by each target selection model as the second probability corresponding to each target selection model.

[0135] like Figure 7 As shown, the attention mechanism module can also include a second normalization unit, which normalizes the sum vector corresponding to the target selection model through the second normalization unit to obtain a normalized vector corresponding to each target selection model, and then processes the normalized vector through the output prediction module in the target selection model to obtain the predicted probability of the target selection model output.

[0136] By normalizing the sum vector, the numerical size of the data is reduced, thereby improving the processing efficiency of the target selection model on the sum vector, thereby improving the efficiency of obtaining the prediction probability.

[0137] In another possible implementation, S360 may also include: normalizing the sum vector corresponding to each target selection model to obtain a normalized vector corresponding to each target selection model; processing the corresponding normalized vector through each target selection model to obtain a first vector corresponding to each target selection model; concatenating the first vector and the intervention vector corresponding to each target selection model to obtain a second vector corresponding to each target selection model; processing the corresponding second vector through each target selection model to obtain the predicted probability output by each target selection model as the second probability corresponding to each target selection model.

[0138] The normalized vector can be processed by the first fully connected network in the target selection model to obtain the processed first vector, and then the first vector corresponding to the target selection model and the intervention vector are spliced ​​to obtain the second vector corresponding to the target selection model, and then the second vector corresponding to the target selection model is processed in sequence by the second fully connected network and the third fully connected network in the target selection model to obtain the output result, and the output result is processed by the sigmoid activation function in the target selection model to obtain the second probability.

[0139] like Figure 7 As shown, the attention mechanism module includes a first fully connected network, and the output prediction module includes: a second splicing unit, a second fully connected network, a third fully connected network and an activation function.

[0140] The normalized vector is processed by the first fully connected network to obtain a first vector, and the first vector and the intervention feature vector are spliced ​​by the second splicing unit to obtain a second vector; the second vector is then input into the second fully connected network, and the output result of the second fully connected network is input into the third fully connected network to obtain the output result of the third fully connected network, and then the output result is processed by the activation function sigmoid to obtain a second probability.

[0141] S370: Obtain a first probability and a second probability corresponding to each target information, and obtain a gain of each target information according to the first probability and the second probability corresponding to each target information.

[0142] S380: Obtain target information with the highest gain among multiple target information as final selection information corresponding to the target user.

[0143] Among them, the description of S370-S380 refers to the description of S140-S150 above and will not be repeated here.

[0144] In this embodiment, the vectors of multiple dimensions included in the user feature vector are spliced, and the spliced ​​vector after splicing is summed with the attention operation result. The sum vector corresponding to the summation result is processed to obtain the output prediction probability as the second probability. The sum vector integrates the vectors of multiple dimensions included in the user feature vector and the attention operation result, so that the sum vector more accurately reflects the probability that the target information is selected by the target user after the intervention is applied, thereby improving the accuracy of the second probability.

[0145] See also Figure 8 , Figure 8 A flowchart of an information selection method proposed in another embodiment of the present application is shown. The method can be applied to an electronic device, which can be Figure 1 The server 10 in the embodiment of the present invention comprises:

[0146] S410 : In response to a target user's access request for a target page, obtain user characteristics of the target user, a plurality of target information, and a target selection model corresponding to an information category of each target information.

[0147] In this application, the target page can refer to a terminal (e.g. Figure 1 20) in the application, for example, the target page may be an information recommendation page of a chat software or a main page of a file management software.

[0148] When the target user sends an operation to open the target page to the terminal, the electronic device receives an access request for the target page based on the target user's operation to open the target page. For example, the target page is the recommended page of chat software D. The target user chats on the chat page of chat software D, then exits the chat page and opens the recommended page. At this time, the electronic device receives an access request based on the target user's operation to open the recommended page.

[0149] S420. Input the user characteristics of the target user into each target selection model to obtain a first probability corresponding to each target selection model. The first probability is used to represent the probability that the target information of the information category corresponding to each target selection model is selected by the target user.

[0150] S430. Input the user characteristics and intervention factors of the target user into the target selection model to obtain the second probability corresponding to each target selection model. The second probability is used to characterize the probability that the target information is selected by the target user when intervention is applied to the target information of the information category corresponding to each target selection model. The intervention is used to increase the exposure of the target information.

[0151] S440: Obtain a first probability and a second probability corresponding to each target information, and obtain a gain of each target information according to the first probability and the second probability corresponding to each target information.

[0152] S450: Obtain target information with the highest gain among multiple target information as final selection information corresponding to the target user.

[0153] Among them, the description of S420-S450 refers to the description of S120-S150 above, and will not be repeated here.

[0154] S460: Display the final selection information on the target page.

[0155] After the final selection information is obtained, the final selection information is displayed on the target page of the terminal.

[0156] As an implementation method, the target user can send an access request through the terminal, and the terminal sends the access request to the server. The server, as the execution body, obtains the final selection information based on the access request, and then the server returns the final selection information to the terminal, and then the terminal outputs the final selection information.

[0157] Optionally, the target page may include at least one display control (also called a resource location, which refers to a location on a specific page where recommended content can be displayed when a user is using an Internet application. For example, when a user is browsing pictures using an application, recommended content can be displayed to the user through the resource location). The final selected information can be displayed through the display control on the target page. Figure 9 As shown, the target page is the main page of the file management software. The main page of the file management software includes a display control 910, and the final selected information is output through the display control. The final selected information is a female dancing video.

[0158] like Figure 9 As shown, the display control also includes a Like button and a Change button. The target user can click the Like button to add the final selected information to the favorite information list. The target user can also click the Change button to trigger a new access request.

[0159] In this embodiment, an access request is sent through the target page, and the final selected information is output on the target page, so that the target user can directly view the displayed information through the target page, thereby improving the user's browsing experience.

[0160] See also Figure 10 , Figure 10 A flowchart of an information selection method proposed in another embodiment of the present application is shown. The method can be applied to an electronic device, which can be Figure 1 The server 10 in the embodiment of the present invention comprises:

[0161] S510: Obtain user characteristics of a target user, multiple target information, and a target selection model corresponding to an information category of each target information.

[0162] S520. Process the user features of the target user through each target selection model to obtain a user feature vector corresponding to each target selection model; process the non-intervention factor through each target selection model to obtain a non-intervention feature vector corresponding to each target selection model.

[0163] Each target selection model may include a feature representation module, which is used to convert the user features of the target user and the non-interference factor into a vector form to obtain a user feature vector and a non-interference feature vector corresponding to each target selection model. The non-interference factor may be 0.

[0164] S530. According to the user feature vector corresponding to each target selection model, obtain the key vector and value vector corresponding to each target selection model; calculate the product of the key vector and the non-interference feature vector corresponding to each target selection model as the non-interference product result corresponding to each target selection model; regularize the non-interference product result corresponding to each target selection model to obtain the non-interference regularized result corresponding to each target selection model; mask the non-interference regularized result corresponding to each target selection model to obtain the non-interference mask processing result corresponding to each target selection model; normalize the non-interference mask processing result corresponding to each target selection model to obtain the non-interference normalized result corresponding to each target selection model; calculate the product of the non-interference normalized result corresponding to each target selection model and the value vector as the non-interference attention operation result corresponding to each target selection model.

[0165] S540. Concatenate the vectors of multiple dimensions included in the user feature vector corresponding to each target selection model to obtain a concatenated vector corresponding to each target selection model; calculate the sum of the concatenated vector corresponding to each target selection model and the result of the non-intervention attention operation as the non-intervention sum vector corresponding to each target selection model; normalize the non-intervention sum vector corresponding to each target selection model to obtain a non-intervention normalized vector corresponding to each target selection model.

[0166] S550. Through each target selection model, the corresponding non-interference normalized vector is processed to obtain the third vector corresponding to each target selection model; the third vector corresponding to each target selection model and the non-interference vector are concatenated to obtain the fourth vector corresponding to each target selection model; through each target selection model, the corresponding fourth vector is processed to obtain the predicted probability output by each target selection model as the first probability.

[0167] Among them, the process of obtaining the first probability in S530-S550 is similar to the calculation process of obtaining the second probability above. The difference is that the intervention factor is replaced by the non-intervention factor in the calculation process, and the intermediate result in the calculation process is also the intermediate result corresponding to the non-intervention factor, which will not be repeated here.

[0168] It can be understood that the non-interference factor is 0, and the process of inputting the non-interference factor and the user characteristics of the target user into the target selection model for processing is actually equivalent to the process of only processing the user characteristics of the target user through the target selection model.

[0169] S560. Input the user characteristics and intervention factors of the target user into the target selection model to obtain the second probability corresponding to each target selection model. The second probability is used to characterize the probability that the target information is selected by the target user when intervention is applied to the target information of the information category corresponding to each target selection model. The intervention is used to increase the exposure of the target information.

[0170] S570: Obtain a first probability and a second probability corresponding to each target information, and obtain a gain of each target information according to the first probability and the second probability corresponding to each target information.

[0171] S580: Obtain target information with the highest gain among multiple target information as final selection information corresponding to the target user.

[0172] Among them, the description of S560-S580 refers to the description of S130-S150 above and is not repeated here.

[0173] In this embodiment, the target selection model can be used to process user characteristics including the target user and non-intervention factors to obtain a first probability, thereby eliminating the need to redeploy a new model. Instead, different values ​​(intervention factors or non-intervention factors) need only be input to obtain the first probability or the second probability, thereby improving the applicability of the target selection model.

[0174] To facilitate understanding of this solution, the information selection method provided by the embodiment of the present application will be described below with reference to an exemplary scenario. In this scenario, the information categories involved include search category, information category, and novel category.

[0175] 1. Preset model training

[0176] like Figure 12 As shown, online data (which may refer to business DAUs, which may include user features used to train the initial model, candidate information for the user, and the selection results corresponding to the candidate information) is used as training samples. First, through a random experiment, the online data is randomly bucketed. 1% of the data in each bucket is selected without any business delivery (no exposure, that is, no intervention). The corresponding data without any business delivery serves as the control group's no-intervention data. Second, 1% of the data is selected from the search category (also called search business), the information category (also called information business), and the novel category (also called novel business). Only a single business content is exposed (daily data only applies intervention to information under any of the search business, information business, and novel business). The data obtained by exposing only a single information category is used as search intervention data, information intervention data, and novel intervention data, respectively.

[0177] At the same time, in order to ensure the effectiveness of the training data and prevent the data of the intervention group (search intervention data, information intervention data, and novel intervention data) from becoming fatigued, the traffic buckets of the random experiment need to be shuffled regularly.

[0178] The initial model is trained based on the search intervention data and the control group's non-intervention data to obtain a preset selection model for the corresponding search category. The initial model is trained based on the information intervention data and the control group's non-intervention data to obtain a preset selection model for the corresponding information category. The initial model is trained based on the novel intervention data and the control group's non-intervention data to obtain a preset selection model for the corresponding novel category.

[0179] At this time, multiple preset selection models are obtained, put online, and deployed on the server.

[0180] 2. Information selection

[0181] like Figure 12 As shown, a user accesses a page through a terminal (the user sends an access request for the target page in the terminal). Based on the access request, the server's master control and decision module retrieves one piece of information from each of the novel content pool, the information content pool, and the search content pool as target information: target information g1 for the novel category, target information g2 for the search category, and target information g3 for the information category. (For each information category, the master control and decision module retrieves multiple pieces of information within that category and selects the one with the highest click-through rate as the target information for that category.)

[0182] At the same time, the master control decision module obtains the user characteristics of the target user according to the access request, and inputs the user characteristics and the intervention factor (the intervention factor can be 1) into the preset selection model of the novel category, the preset selection model of the information category, and the preset selection model of the search category, respectively, to obtain the second probability of the target information of the novel category, the second probability of the target information of the information category, and the second probability of the target information of the search category.

[0183] The preset selection model for novel categories processes user characteristics and intervention factors as follows:

[0184] like Figure 13 As shown, the user features (including d-dimensional features, d is a non-zero natural number, are represented as 、 、…、 ) Input the feature representation module to obtain the user feature vector, which is also a d-dimensional vector, and input the intervention factor t into the feature representation module to obtain the intervention feature vector.

[0185] In the attention mechanism module, an attention mechanism operation is performed on the user feature vector and the intervention feature vector to obtain an attention operation result; the attention mechanism module splices the d-dimensional vectors included in the user feature vector to obtain a spliced ​​vector, and then the attention mechanism module first adds the spliced ​​vector and the attention operation result, and then normalizes the added result (that is, the sum vector of the above embodiment) to obtain a normalized vector, and then the first fully connected network in the attention mechanism module processes the normalized vector to obtain a first vector.

[0186] In the output prediction module, the intervention feature vector and the first vector are concatenated to obtain a second vector, which is then sequentially input into the second fully connected network, the third fully connected network, and the activation function in the output prediction module to obtain a predicted probability, which serves as the second probability corresponding to the preset selection model of the novel category.

[0187] The processing process of user characteristics and intervention factors by the preset selection model of the information category and the preset selection model of the search category refers to the processing process of user characteristics and intervention factors by the preset selection model of the novel category, and will not be repeated here.

[0188] The preset selection model for novel categories processes user characteristics and non-intervention factors as follows:

[0189] The user features are input into the feature representation module to obtain the user feature vector, which is also a d-dimensional vector. The non-intervention factor is input into the feature representation module to obtain the non-intervention feature vector.

[0190] In the attention mechanism module, an attention mechanism operation is performed on the user feature vector and the non-intervention feature vector to obtain a non-intervention attention operation result; the attention mechanism module splices the d-dimensional vector included in the user feature vector to obtain a spliced ​​vector, and then the attention mechanism module first adds the spliced ​​vector and the non-intervention attention operation result, and then normalizes the added result (that is, the non-intervention sum vector of the above embodiment) to obtain a non-intervention normalized vector, and then the first fully connected network in the attention mechanism module processes the non-intervention normalized vector to obtain a third vector.

[0191] In the output prediction module, the non-intervention feature vector and the third vector are spliced ​​to obtain a fourth vector, and then the fourth vector is sequentially input into the second fully connected network, the third fully connected network and the activation function in the output prediction module to obtain the predicted probability as the first probability corresponding to the preset selection model of the novel category.

[0192] The processing process of user characteristics and non-intervention factors by the preset selection model of the information category and the preset selection model of the search category can be referred to the processing process of user characteristics and non-intervention factors by the preset selection model of the novel category, and will not be repeated here.

[0193] According to the first probability and the second probability corresponding to the target information g1, the gain h1 of the target information g1 is determined; according to the first probability and the second probability corresponding to the target information g2, the gain h2 of the target information g2 is determined; according to the first probability and the second probability corresponding to the target information g3, the gain h3 of the target information g3 is determined.

[0194] The target information g3 corresponding to h3 with the highest gain is obtained as the final selected information, and the target information g3 is information of the information category.

[0195] like Figure 12 As shown, after the general control decision module of the server obtains the target information g3, it sends the target information g3 to the terminal through the resource position background corresponding to the target information g3 (different categories of resource positions can correspond to different resource position backgrounds. For example, the resource position background of the novel category and the resource position background of the information category can be different. The resource position is also the display control of the above embodiment), and the terminal outputs the target information g3 in the resource position of the target page.

[0196] The possibility of the output final target information g3 being selected by the target user is the highest among the target information g1, the target information g2 and the target information g3.

[0197] This solution is centered on causal effects. From the perspective of causal inference, it models the user's response when different businesses (information categories in the above embodiments) are intervened. This allows for a more scientific measurement of the DAU increase brought about when users select information under a business. Finally, the estimated CTR of the business information is combined to jointly determine the information gain for user access, thereby maximizing the overall daily active users of each business.

[0198] See also Figure 14 , Figure 14 A block diagram of an information selection device proposed in one embodiment of the present application is shown. The device 1100 includes:

[0199] The first acquisition module 1110 is used to obtain the user characteristics of the target user, multiple target information, and a target selection model corresponding to the information category of each target information;

[0200] A first obtaining module 1120 is configured to input the user characteristics of the target user into each target selection model to obtain a first probability corresponding to each target selection model, wherein the first probability is used to represent the probability that the target information of the information category corresponding to each target selection model is selected by the target user;

[0201] A second obtaining module 1130 is configured to input the user characteristics of the target user and the intervention factor into the target selection model to obtain a second probability corresponding to each target selection model. The second probability is used to represent the probability that the target information will be selected by the target user when an intervention is applied to the target information of the information category corresponding to each target selection model. The intervention is used to increase the exposure of the target information.

[0202] The second acquisition module 1140 is configured to acquire a first probability and a second probability corresponding to each target information, and obtain a gain of each target information according to the first probability and the second probability corresponding to each target information;

[0203] The information acquisition module 1150 is used to acquire target information with the highest gain from multiple target information as final selection information corresponding to the target user.

[0204] Optionally, the second acquisition module 1140 is also used to calculate the difference between the second probability corresponding to each target information and the first probability as the probability difference of each target information; obtain the estimated click-through rate corresponding to each target information; calculate the product of the probability difference corresponding to each target information and the estimated click-through rate as the gain of each target information.

[0205] Optionally, the second acquisition module 1130 is also used to process the user features of the target user through each target selection model to obtain the user feature vector corresponding to each target selection model; process the intervention factor through each target selection model to obtain the intervention feature vector corresponding to each target selection model; perform attention mechanism operation on the user feature vector and intervention feature vector corresponding to each target selection model to obtain the attention operation result corresponding to each target selection model; process the corresponding user feature vector and attention operation result through each target selection model to obtain the predicted probability output by each target selection model as the second probability corresponding to each target selection model.

[0206] Optionally, the second acquisition module 1130 is also used to obtain the key vector and value vector corresponding to each target selection model based on the user feature vector corresponding to each target selection model; calculate the product of the key vector and the intervention feature vector corresponding to each target selection model as the product result corresponding to each target selection model; regularize the product result corresponding to each target selection model to obtain the regularized result corresponding to each target selection model; mask the regularized result corresponding to each target selection model to obtain the mask processing result corresponding to each target selection model; normalize the mask processing result corresponding to each target selection model to obtain the normalized result corresponding to each target selection model; calculate the product of the normalized result corresponding to each target selection model and the value vector as the attention operation result corresponding to each target selection model.

[0207] Optionally, the second acquisition module 1130 is also used to splice the vectors of multiple dimensions included in the user feature vector corresponding to each target selection model to obtain the splicing vector corresponding to each target selection model; calculate the sum of the splicing vector corresponding to each target selection model and the attention operation result as the sum vector corresponding to each target selection model; and process the corresponding sum vector through each target selection model to obtain the predicted probability output by each target selection model.

[0208] Optionally, the second acquisition module 1130 is also used to normalize the sum vector corresponding to each target selection model to obtain a normalized vector corresponding to each target selection model; through each target selection model, the corresponding normalized vector is processed to obtain the predicted probability output by each target selection model.

[0209] Optionally, the second acquisition module 1130 is also used to process the corresponding normalized vectors through each target selection model to obtain the first vector corresponding to each target selection model; concatenate the first vector and the intervention vector corresponding to each target selection model to obtain the second vector corresponding to each target selection model; and process the corresponding second vector through each target selection model to obtain the predicted probability output by each target selection model.

[0210] Optionally, the device also includes a display module; the first acquisition module 1110 is also used to obtain the user characteristics of the target user, multiple target information and the target selection model corresponding to the information category of each target information in response to the target user's access request for the target page; the display module is used to display the final selected information on the target page.

[0211] Optionally, the first acquisition module 1110 is further used to determine the target selection model corresponding to each target information from multiple preset selection models according to the information category of each target information, and the multiple preset selection models correspond to different information categories respectively.

[0212] Optionally, the device also includes a training module for obtaining training samples corresponding to each information category. The training samples corresponding to each information category include user characteristics of the first user, user characteristics of the second user, a first selection result corresponding to the first sample information under the information category, and a second selection result corresponding to the second sample information under the information category. The first selection result refers to the selection result of the first user for the first sample information when no intervention is applied to the first sample information, and the second selection result refers to the selection result of the second user for the second sample information when intervention is applied to the second sample information. The initial model is trained according to the training samples corresponding to each information category to obtain a target selection model corresponding to each information category.

[0213] It should be noted that the device embodiment in this application corresponds to the aforementioned method embodiment. The specific principles in the device embodiment can be found in the contents of the aforementioned method embodiment and will not be repeated here.

[0214] Figure 15 A structural block diagram of an electronic device for executing the information selection method according to an embodiment of the present application is shown. The electronic device may be Figure 1 In the server 10, it should be noted that, Figure 15 The computer system 1200 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0215] like Figure 15 As shown, computer system 1200 includes a central processing unit (CPU) 1201, which can perform various appropriate actions and processes, such as executing the methods described in the above embodiments, based on programs stored in read-only memory (ROM) 1202 or programs loaded from storage 1208 into random access memory (RAM) 1203. RAM 1203 also stores various programs and data required for system operation. CPU 1201, ROM 1202, and RAM 1203 are interconnected via bus 1204. An input / output (I / O) interface 1205 is also connected to bus 1204.

[0216] The following components are connected to the I / O interface 1205: an input section 1206 including a keyboard, mouse, and the like; an output section 1207 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and speakers; a storage section 1208 including a hard disk; and a communication section 1209 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to the I / O interface 1205 as needed. Removable media 1211, such as a magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, is installed in the drive 1210 as needed, allowing computer programs read from the media to be installed in the storage section 1208 as needed.

[0217] In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for executing the methods illustrated in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section 1209 and / or installed from removable media 1211. When executed by the central processing unit (CPU) 1201, the computer program performs the various functions defined in the system of the present application.

[0218] It should be noted that the computer-readable medium described in the embodiments of this application may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal transmitted in baseband or as part of a carrier wave, which carries 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.

[0219] 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.

[0220] 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.

[0221] 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.

[0222] 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 an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the method of any of the above embodiments.

[0223] 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.

[0224] 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 a number of instructions to enable an electronic 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.

[0225] Those skilled in the art will readily appreciate 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 the present application and include common knowledge or customary techniques in the art that are not disclosed in this application. It should be understood that this application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of this application is limited only by the appended claims.

[0226] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An information selection method, characterized in that: The method comprises: Obtaining user characteristics of a target user, multiple target information, and a target selection model corresponding to an information category of each target information; Inputting the user characteristics of the target user into each of the target selection models to obtain a first probability corresponding to each of the target selection models, wherein the first probability is used to represent the probability that the target information of the information category corresponding to each of the target selection models is selected by the target user; Processing the user features of the target user by each target selection model to obtain a user feature vector corresponding to each target selection model; Processing the intervention factor through each of the target selection models to obtain an intervention feature vector corresponding to each of the target selection models; Performing an attention mechanism operation on the user feature vector and the intervention feature vector corresponding to each target selection model to obtain an attention operation result corresponding to each target selection model; By processing the corresponding user feature vector and the attention calculation result of each target selection model, a second probability corresponding to each target selection model is obtained; the second probability is used to represent the probability that the target information will be selected by the target user when an intervention is applied to the target information of the information category corresponding to each target selection model, wherein the intervention is used to increase the exposure of the target information; Obtaining a first probability and a second probability corresponding to each piece of target information, and obtaining a gain of each piece of target information according to the first probability and the second probability corresponding to each piece of target information; The target information with the highest gain among the multiple target information is obtained as the final selected information corresponding to the target user.

2. The method according to claim 1, characterized in that Obtaining a gain for each target information according to the first probability and the second probability corresponding to each target information includes: Calculating a difference between the second probability and the first probability corresponding to each piece of target information as a probability difference for each piece of target information; Obtaining an estimated click-through rate corresponding to each target information; The product of the probability difference corresponding to each target information and the estimated click rate is calculated as the gain of each target information.

3. The method according to claim 1, characterized in that The performing of an attention mechanism operation on the user feature vector and the intervention feature vector corresponding to each of the target selection models to obtain an attention operation result corresponding to each of the target selection models includes: Obtaining a key vector and a value vector corresponding to each target selection model according to the user feature vector corresponding to each target selection model; Calculating the product of the key vector and the intervention feature vector corresponding to each of the target selection models as the product result corresponding to each of the target selection models; Regularizing the product results corresponding to each of the target selection models to obtain regularized results corresponding to each of the target selection models; Performing mask processing on the regularization results corresponding to each target selection model to obtain the mask processing results corresponding to each target selection model; Normalizing the mask processing results corresponding to each target selection model to obtain a normalized result corresponding to each target selection model; Calculate the product of the normalized result corresponding to each target selection model and the value vector as the attention operation result corresponding to each target selection model.

4. The method according to claim 1, wherein The step of processing the corresponding user feature vector and the attention operation result by each target selection model to obtain the second probability corresponding to each target selection model includes: Performing splicing processing on vectors of multiple dimensions included in the user feature vector corresponding to each target selection model to obtain a splicing vector corresponding to each target selection model; Calculate the sum of the concatenation vector corresponding to each target selection model and the attention operation result as the sum vector corresponding to each target selection model; Through each of the target selection models, the corresponding sum vectors are processed to obtain the second probability corresponding to each of the target selection models.

5. The method according to claim 4, characterized in that The step of processing the corresponding sum vectors by each target selection model to obtain the second probability corresponding to each target selection model includes: Normalizing the sum vector corresponding to each target selection model to obtain a normalized vector corresponding to each target selection model; By processing the corresponding normalized vectors through each of the target selection models, the second probability corresponding to each of the target selection models is obtained.

6. The method according to claim 5, characterized in that The step of processing the corresponding normalized vectors by each target selection model to obtain the second probability corresponding to each target selection model includes: Processing the corresponding normalized vectors of each target selection model to obtain a first vector corresponding to each target selection model; performing concatenation processing on the first vector and the intervention vector corresponding to each of the target selection models to obtain a second vector corresponding to each of the target selection models; By processing the second vectors corresponding to each target selection model, the second probabilities corresponding to each target selection model are obtained.

7. The method according to claim 1, characterized in that The step of obtaining the user characteristics of the target user, multiple target information, and a target selection model corresponding to the information category of each target information includes: In response to an access request from the target user for a target page, obtaining a user feature of the target user, the plurality of target information, and a target selection model corresponding to an information category of each of the target information; After obtaining the information with the highest gain among the plurality of target information as the final selected information corresponding to the target user, the method further includes: The final selection information is displayed on the target page.

8. The method according to claim 1, characterized in that The acquiring of a target selection model corresponding to the information category of each target information includes: According to the information category of each target information, a target selection model corresponding to each target information is determined from a plurality of preset selection models, and the plurality of preset selection models respectively correspond to different information categories.

9. The method according to any one of claims 1 to 8, characterized in that The training method of the target selection model includes: Obtaining training samples corresponding to each of the information categories, where the training samples corresponding to each of the information categories include user features of a first user, user features of a second user, a first selection result corresponding to first sample information under the information category, and a second selection result corresponding to second sample information under the information category, where the first selection result refers to a selection result of the first user for the first sample information when no intervention is applied to the first sample information, and the second selection result refers to a selection result of the second user for the second sample information when intervention is applied to the second sample information; The initial model is trained according to the training samples corresponding to each of the information categories to obtain a target selection model corresponding to each of the information categories.

10. An information selection device, characterized in that: The device comprises: A first acquisition module is used to acquire user characteristics of a target user, multiple target information, and a target selection model corresponding to an information category of each target information; a first obtaining module, configured to input the user characteristics of the target user into each of the target selection models, and obtain a first probability corresponding to each of the target selection models, wherein the first probability is used to represent the probability that the target information of the information category corresponding to each of the target selection models is selected by the target user; A second acquisition module is configured to process the user features of the target user through each target selection model to obtain a user feature vector corresponding to each target selection model; process the intervention factor through each target selection model to obtain an intervention feature vector corresponding to each target selection model; perform an attention mechanism operation on the user feature vector and the intervention feature vector corresponding to each target selection model to obtain an attention operation result corresponding to each target selection model; process the corresponding user feature vector and the attention operation result through each target selection model to obtain a second probability corresponding to each target selection model; the second probability is used to represent the probability that the target information is selected by the target user when an intervention is applied to the target information of the information category corresponding to each target selection model, and the intervention is used to increase the exposure of the target information; A second acquisition module is configured to acquire a first probability and a second probability corresponding to each piece of target information, and obtain a gain of each piece of target information according to the first probability and the second probability corresponding to each piece of target information; The information acquisition module is used to acquire the target information with the highest gain from the multiple target information as the final selection information corresponding to the target user.

11. An electronic device, characterized in that: include: one or more processors; Memory; One or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs being configured to execute the method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, which can be called by a processor to execute the method according to any one of claims 1 to 9.

Images