Data recommendation method and related device

Abstract

The invention provides a data recommendation method and a related device. The embodiment of the invention can be applied to the field of artificial intelligence. The method comprises the steps of obtaining a data recommendation request, obtaining a feature vector of a target behavior, a feature vector of a target file and a feature vector of a to-be-recommended file, determining a time coefficient according to the target behavior, determining a time sequence of accessing the target file by a target object according to the feature vector of the target file and the time coefficient, and generating a first vector, inputting the first vector and the feature vector of the target file into a neural network to obtain a weight coefficient of the to-be-recommended file, fusing the weight coefficient of the to-be-recommended file and the feature vector of the to-be-recommended file, inputting the fused weight coefficient and feature vector into the full connection layer, and obtaining the click passing rate of the to-be-recommended file. The product between the feature vector and the time coefficient of the target file is calculated in the Embedding layer of the model, and the influence of the time when the target object accesses the target file on the click passing rate prediction is introduced, so that the prediction accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular, to a method for data recommendation and related devices. Background Art

[0002] With the continuous development of artificial intelligence technology, artificial intelligence technology has penetrated into all aspects of life. In the field of product recommendation, the clickthrough rate (CTR) of a product can be predicted by using the information in the user's historical behavior data.

[0003] Although the user's historical behavior data can represent the user's interest direction, with the passage of time, the user's interest direction also changes continuously. The user's historical behavior data should have different influences when predicting the CTR of a product. In the process of predicting the CTR of a product, the gated recurrent unit (GRU) module in natural language processing can be used to realize the influence of time in the user's historical behavior data on the prediction result. However, since the GRU module is a serial computing module, applying the GRU module in the model will cause the time required for model inference to increase exponentially.

[0004] How to improve the influence of time in the user's historical behavior data on the preset result while ensuring the model calculation speed has become an urgent problem to be solved currently. Summary of the Invention

[0005] Embodiments of this application provide a method for data recommendation and related devices, which are used to improve the influence of time information in the user's historical behavior data on the preset result while ensuring the model calculation speed.

[0006] The first aspect of this application provides a method for data recommendation, including:

[0007] Obtain a data recommendation request, where the data recommendation request carries the identifier of the target object;

[0008] Obtain the target behavior, the target file, and the file to be recommended according to the data recommendation request, where the target behavior is used to describe the access of the target object to the target file;

[0009] Determine a time coefficient according to the target behavior, where the time coefficient is used to describe the time series of the target object accessing the target file;

[0010] Perform feature mapping on the target behavior, the target file, and the file to be recommended to obtain the feature vector of the target behavior, the feature vector of the target file, and the feature vector of the file to be recommended;

[0011] Generate a first vector according to the feature vector of the target file and the time coefficient;

[0012] Input the first vector and the feature vector of the file to be recommended into a neural network to obtain the weight coefficient of the file to be recommended;

[0013] After fusing the weight coefficient of the file to be recommended and the feature vector of the file to be recommended, input them into a fully connected layer to obtain the click-through rate of the file to be recommended;

[0014] Recommend the file to be recommended whose click-through rate meets the preset conditions to the target object.

[0015] The second aspect of this application provides a data recommendation device, including:

[0016] An acquisition unit for acquiring a data recommendation request, and the data recommendation request carries the identifier of the target object;

[0017] The acquisition unit is further configured to acquire the target behavior, the target file and the file to be recommended according to the data recommendation request, and the target behavior is used to describe the access of the target object to the target file;

[0018] A calculation unit for determining a time coefficient according to the target behavior, and the time coefficient is used to describe the time series of the target object accessing the target file;

[0019] A generation unit for performing feature mapping on the target behavior, the target file and the file to be recommended to obtain the feature vector of the target behavior, the feature vector of the target file and the feature vector of the file to be recommended;

[0020] The generation unit is further configured to generate a first vector according to the feature vector of the target file and the time coefficient;

[0021] The calculation unit is further configured to input the first vector and the feature vector of the file to be recommended into a neural network to obtain the weight coefficient of the file to be recommended;

[0022] The calculation unit is further configured to input the weight coefficient of the file to be recommended and the feature vector of the file to be recommended into a fully connected layer after fusion to obtain the click-through rate of the file to be recommended;

[0023] A recommendation unit for recommending the file to be recommended whose click-through rate meets the preset conditions to the target object.

[0024] In a possible implementation manner of the second aspect, the target behavior carries the occurrence time of the target object's access to the target file;

[0025] The calculation unit is specifically configured to determine the time coefficient according to the occurrence time of the target object's access to the target file.

[0026] In a possible implementation manner of the second aspect, the target behavior includes H accesses of the target object to the target file, and H is a positive integer;

[0027] A calculation unit, specifically configured to:

[0028] According to the target behavior, determine the proportion of the number of unaccessed access times at the moment when the target object accesses the first target file, and obtain the time coefficient to be processed, where the first target file is included in the target file;

[0029] Perform an exponential calculation on the time coefficient to be processed to obtain the time coefficient.

[0030] In a possible implementation manner of the second aspect, the calculation unit is specifically configured to:

[0031] Process the first vector and the feature vector of the file to be recommended to obtain the outer product of the first vector and the feature vector of the file to be recommended;

[0032] Concatenate the first vector, the feature vector of the file to be recommended, and the outer product of the first vector and the feature vector of the file to be recommended to obtain a concatenated vector;

[0033] Input the concatenated vector into an activation function to obtain the weight coefficient of the file to be recommended.

[0034] In a possible implementation manner of the second aspect, the file to be recommended includes at least two files to be recommended;

[0035] The calculation unit is specifically configured to:

[0036] Fuse the weight coefficients of at least two files to be recommended with the feature vectors of the files to be recommended respectively to obtain at least two target feature vectors;

[0037] After concatenating at least two target feature vectors, input them into a fully connected layer to obtain the click-through rate of the file to be recommended.

[0038] In a possible implementation manner of the second aspect, the acquisition unit is further configured to acquire the feature vector of the target object and the scene feature vector of the target object;

[0039] The calculation unit is specifically configured to concatenate at least two target feature vectors, the feature vector of the target object, and the scene feature vector of the target object, and then input them into a fully connected layer to obtain the click-through rate of the file to be recommended.

[0040] In a possible implementation manner of the second aspect, the target behavior is used to describe the browsing of the target object for the target file, or the acquisition of the target object for the target file, or the browsing of the target object for the target file and the acquisition of the target object for the target file.

[0041] Another aspect of the present application provides a computer device, including:

[0042] A memory, a transceiver, a processor, and a bus system;

[0043] The memory is used to store programs;

[0044] The processor is used to execute the programs in the memory, including executing the methods in the above aspects;

[0045] The bus system is used to connect the memory and the processor, so that the memory and the processor can communicate.

[0046] Another aspect of the present application provides a computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When it runs on a computer, it causes the computer to execute the methods in the above aspects.

[0047] Another aspect of the present application provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions. The computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in the above aspects.

[0048] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0049] The present application provides a method and a related device for data recommendation. By obtaining a data recommendation request, the data recommendation request carries the identifier of the target object, obtaining a target behavior, a target file, and a file to be recommended according to the data recommendation request, the target behavior is used to describe the access of the target object to the target file, and determining a time coefficient according to the target behavior, the time coefficient is used to describe the time series of the target object accessing the target file, extracting features of the target behavior, the target file, and the file to be recommended, obtaining a feature vector of the target behavior, a feature vector of the target file, and a feature vector of the file to be recommended, generating a first vector according to the feature vector of the target file and the time coefficient, inputting the first vector and the feature vector of the file to be recommended into a neural network, obtaining a weight coefficient of the file to be recommended, and after fusing the weight coefficient of the file to be recommended with the feature vector of the file to be recommended, inputting it into a fully connected layer to obtain the click-through rate of the file to be recommended. Recommending the file to be recommended whose click-through rate meets the preset conditions to the target object. By calculating the product between the feature vector of the target file and the time coefficient in the Embedding layer of the model, the influence of the time when the target object accesses the target file on the click-through rate prediction is introduced, and without a large amount of time consumption, the influence of time on the prediction result in the user's historical behavior data is improved, thereby improving the prediction accuracy. Description of the Drawings

[0050] Figure 1A schematic architecture diagram of the data recommendation system provided by the embodiments of the present application;

[0051] Figure 2 A schematic flowchart of the data recommendation method provided by the embodiments of the present application;

[0052] Figure 3a A schematic diagram of a target behavior provided by the embodiments of the present application;

[0053] Figure 3b Another schematic diagram of a target behavior provided by the embodiments of the present application;

[0054] Figure 3c Another schematic diagram of a target behavior provided by the embodiments of the present application;

[0055] Figure 4 A schematic diagram of an attention neural network model provided by the embodiments of the present application;

[0056] Figure 5a A schematic diagram of a click-through rate prediction model provided by the embodiments of the present application;

[0057] Figure 5b Another schematic diagram of a click-through rate prediction model provided by the embodiments of the present application;

[0058] Figure 6 A schematic structural diagram of the data recommendation device provided by the embodiments of the present application;

[0059] Figure 7 A schematic structural diagram of the server provided by the embodiments of the present application. Detailed implementation manners

[0060] The embodiments of the present application provide a data recommendation method and related devices, which are used to improve the influence of time information in user historical behavior data on preset results while ensuring the model calculation speed.

[0061] The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and drawings of the present application are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "correspond to" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0062] Artificial Intelligence (AI) uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, including theories, methods, technologies, and application systems for perceiving the environment, acquiring knowledge, and using knowledge to achieve optimal results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines to enable machines to have the functions of perception, reasoning, and decision-making.

[0063] Artificial intelligence technology is an interdisciplinary subject with a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0064] Machine Learning (ML) is an interdisciplinary subject across multiple fields, involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning.

[0065] To facilitate the understanding of the technical solutions provided by the embodiments of this application, some key terms used in the embodiments of this application are first explained here:

[0066] The deep interest network (DIN) is a deep interest network theory for CTR prediction in the e-commerce field by Alibaba. Combining the attention mechanism, it fully explores the information in the user's historical behavior data.

[0067] Deep Interest Evolution Network (DIEN) takes into account user interests, proposing that user interests are diverse. It uses an attention mechanism to capture the relative interest in the target event and applies this adaptive interest representation to model prediction. However, most models of this type directly regard user behavior as interests, while the latent interests of users are often difficult to fully represent through behavior. Therefore, it is necessary to explore the true interests of users behind their behavior and consider the dynamic changes in user interests.

[0068] CTR, that is, click-through rate, is a commonly used term in Internet advertising, referring to the click-through rate of online advertisements (such as image ads, text ads, keyword ads, ranking ads, video ads, etc.), that is, the actual number of clicks on the ad (strictly speaking, it can be the number of arrivals at the target page) divided by the display volume of the ad (Show content).

[0069] An embedding vector is the mapping result obtained by mapping high-dimensional raw data (such as images, sentences, etc.) to a low-dimensional manifold, making the high-dimensional raw data separable after being mapped to the low-dimensional manifold.

[0070] Sequential features: Sequential features are multi-valued features of user conversion behaviors such as clicks. They are sequences composed of the identity identifiers (identify document, ID) of user conversions in chronological order. For example, the occurrence times of event 1, event 2, and event 3 are sequential features of user conversions from near to far on the time axis.

[0071] Position-embedding: It is a method for encoding position information in sequential data. When processing sequential data (such as natural language text or time series data), capturing the order relationship of elements is very important for understanding the context. Position embedding allows the model to identify the position of elements in the sequence and their relative relationships.

[0072] With the continuous development of artificial intelligence technology, artificial intelligence technology has penetrated into all aspects of life. In the field of product recommendation, the DIN model can recommend products / content that users are interested in by predicting the click-through rate (CTR) of products / content.

[0073] However, the DIN model does not consider the relationship between the time when the user's historical behavior data occurs and the product CTR, treats all the user's historical behavior data in the behavior sequence equally, and processes it in the same way. Although the DIEN model takes into account the relationship between the time when the user's historical behavior data occurs and the product CTR, and although the DIEN model introduces the GRU module in natural language processing, since the GRU module is a serial computing module, applying the GRU module in the model will cause the time required for model inference to increase exponentially.

[0074] How to improve the impact of time in the user's historical behavior data on the prediction result while ensuring the model calculation speed has become an urgent problem to be solved.

[0075] To address the above problems, this application proposes that by obtaining a data recommendation request carrying the identifier of the target object, acquiring the target behavior, the target file, and the file to be recommended according to the data recommendation request, where the target behavior is used to describe the access of the target object to the target file, and determining the time coefficient according to the target behavior, the time coefficient is used to describe the time series of the target object accessing the target file, extracting features from the target behavior, the target file, and the file to be recommended to obtain the feature vector of the target behavior, the feature vector of the target file, and the feature vector of the file to be recommended, generating a first vector according to the feature vector of the target file and the time coefficient, inputting the first vector and the feature vector of the file to be recommended into a neural network to obtain the weight coefficient of the file to be recommended, and after fusing the weight coefficient of the file to be recommended with the feature vector of the file to be recommended, inputting it into a fully connected layer to obtain the click-through rate of the file to be recommended. Recommend the file to be recommended whose click-through rate meets the preset conditions to the target object. By calculating the product between the feature vector of the target file and the time coefficient in the Embedding layer of the model, the impact of the time when the target object accesses the target file on click-through rate prediction is introduced, and without a large amount of time consumption, the influence of time on the prediction result in the user's historical behavior data is improved, thereby improving the prediction accuracy.

[0076] For ease of understanding, please refer to Figure 1 , Figure 1 which is the application environment diagram of the data recommendation method in the embodiments of this application, as Figure 1As shown in the figure, in the embodiment of the present application, the method for data recommendation is applied to a data recommendation system. The data recommendation system includes: a server and a terminal device; among them, the server can be an independent physical server, or a server cluster or distributed system composed 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, Content Delivery Network (CDN), and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the server can be directly or indirectly connected through wired or wireless communication methods, and the embodiments of the present application do not limit this here.

[0077] The server first obtains a data recommendation request, and the data recommendation request carries the identifier of the target object;

[0078] According to the data recommendation request, obtain the target behavior, the target file, and the file to be recommended. The target behavior is used to describe the access of the target object to the target file;

[0079] Determine the time coefficient according to the target behavior. The time coefficient is used to describe the time series of the target object accessing the target file;

[0080] Perform feature mapping on the target behavior, the target file, and the file to be recommended to obtain the feature vector of the target behavior, the feature vector of the target file, and the feature vector of the file to be recommended;

[0081] Generate a first vector according to the feature vector of the target file and the time coefficient;

[0082] Input the first vector and the feature vector of the file to be recommended into a neural network to obtain the weight coefficient of the file to be recommended;

[0083] After fusing the weight coefficient of the file to be recommended and the feature vector of the file to be recommended, input it into a fully connected layer to obtain the click-through rate of the file to be recommended;

[0084] Recommend the file to be recommended whose click-through rate meets the preset conditions to the target object.

[0085] Next, from the perspective of the server, the method for data recommendation in the present application will be introduced. Please refer to Figure 2 , the method for data recommendation provided by the embodiment of the present application includes: step S101 to step S108. Specifically:

[0086] S101. Obtain a data recommendation request;

[0087] Obtain a data recommendation request, where the data recommendation request carries the ID of the target object.

[0088] Exemplarily, the data recommendation request can be obtained by responding to the input of the ID of the target object, or the web page can periodically generate the data recommendation request according to the logged-in target object, and there is no limitation here.

[0089] S102. Obtain the target behavior, target file, and files to be recommended according to the data recommendation request;

[0090] Among them, the target behavior is used to describe the access of the target object to the target file, and the files to be recommended are files preset by the target user, and there is no limitation on the specific file form.

[0091] Specifically, the target behavior can include the access of the target object to the target file within a preset duration in the past; the target behavior can also carry the time when the access of the target object to the target file occurs, that is, the access timestamp of the target object to the target file.

[0092] For example, the target behavior can be the click sequence generated by the target object accessing the target object within a preset duration in the past, and the length of the click sequence is H, that is, the target object has accessed the target file H times within a preset duration in the past, and H is a positive integer; at the same time, the access timestamp of the target object to the target file can also be carried in the click sequence.

[0093] Please refer to Figure 3a , Figure 3b and Figure 3c In some possible application scenarios, the record of the target behavior can be as Figure 3a shown, carrying the identity identifier (identify document, ID) of the target file to mark the access of the target object to different target files.

[0094] In some possible application scenarios, the record of the target behavior can be as Figure 3b shown, carrying the access timestamp of the target object to the target file and the identity identifier ID of the target file, so as to record that the target object has accessed a certain target file at a certain moment.

[0095] In some possible application scenarios, the record of the target behavior can be as Figure 3cAs shown, it carries the access timestamp of the target object for the target file, the identity ID of the target file, and the first identifier or the second identifier used to mark the access type of the target object for the target file. It is used to record the access of the target object to a certain target file in the form of acquisition, or the access of the target object to a certain target file in the form of browsing. Among them, the first identifier is used to mark the access of the target object to the target file as browsing, and the second identifier is used for the access of the target object to the target file as acquisition.

[0096] It can be understood that the description of the target behavior here is only an example. In actual applications, other forms can also be used to record the target behavior. When specifically setting, it should be combined with the specific application scenario, and there is no limit here.

[0097] In Figure 3c In the application scenario introduced as above, according to the requirements, the target behavior can describe the browsing of the target object for the target file, or the acquisition of the target object for the target file, or the browsing of the target object for the target file and the acquisition of the target object for the target file. There is no limit here.

[0098] In some specific implementation cases, when the target behavior includes a single access type such as the browsing of the target object for the target file or the acquisition of the target object for the target file, the solution provided in the embodiments of the present application can be executed.

[0099] When the target behavior includes both the browsing of the target object for the target file and the acquisition of the target object for the target file at the same time, the first weight can be given to the browsing of the target object for the target file, and the second weight can be given to the acquisition of the target object for the target file. The influence on the file to be recommended when the target behavior is different can be controlled by adjusting the first weight and the second weight. There is no limit here.

[0100] S103. Determine the time coefficient according to the target behavior;

[0101] Among them, the time coefficient is used to describe the time series of the target object accessing the target file.

[0102] There are various ways to determine the time coefficient according to the target behavior, which will be introduced separately below:

[0103] 1) Determine the time coefficient according to the occurrence order of the target behavior;

[0104] Determine the time coefficient corresponding to the target behavior according to the occurrence order of the target behavior.

[0105] For example, if H target behaviors are collected, and pos(i) indicates the occurrence order of the target object's access to the first target file among the H target behaviors, where pos(i) is a natural number less than H, the first target file is included in the target file, and H is a positive integer, then the time coefficient can be:

[0106] β = pos(i);

[0107] 2) Determine the time coefficient according to the occurrence time of the target behavior;

[0108] For example, when the click sequence length is H, pos(i) is the position of the target object's access to the first target file among the H accesses, where pos(i) is a natural number less than H, the first target file is included in the target file, and H is a positive integer, then the time coefficient is:

[0109] β = pos(i);

[0110] Or,

[0111] where γ is the attenuation hyperparameter, and the larger the value of γ, the faster the attenuation of the time coefficient.

[0112] In a possible scenario, when the obtained target behavior is as shown in Figure 3b or Figure 3c the target behavior shown, the time coefficient can also be determined according to the occurrence time of the target object's access to the target file. Among them, the occurrence time of the target object's access to the target file can be recorded in the obtained target behavior in the form of a timestamp, and there is no limitation here.

[0113] When the target behavior includes H accesses of the target object to the target file, sort the access times of the target object to the target file in descending order according to the timestamps in the target behavior. pos(i) is the position of the target object's access to the first target file among the H accesses, that is, the order of the target object's access to the first target file in the sorting.

[0114] Sort the access times of the target object to the target file in descending order according to the timestamps in the target behavior.

[0115] 3) Determine the time coefficient according to the type of the target behavior combined with the occurrence time of the target behavior.

[0116] In a possible scenario, the obtained target behavior is as Figure 3cWhen the target behavior shown in the figure is involved, the target behavior can be screened according to requirements by using the first identifier and the second identifier. For example, in a scenario where it is more desired to analyze the acquisition of the target file by the target object, among the N target behaviors obtained, the target behaviors with the screening identifier being the second identifier are screened, and the behaviors of H target users for acquiring the target file are obtained. Then the time coefficient is:

[0117]

[0118] It can be understood that the description of the determination method of the time coefficient here is only an example. In actual applications, it should be set according to the specific scenario, and no restrictions are imposed here.

[0119] Among them, γ is the decay hyperparameter, and the larger the value of γ, the faster the decay of the time coefficient.

[0120] It should be noted that there is no clear sequence between step S103 and step S104. In actual applications, it should be set according to the specific usage scenario, and no restrictions are imposed here.

[0121] S104. Perform feature mapping on the target behavior, the target file, and the file to be recommended to obtain the feature vector of the target behavior, the feature vector of the target file, and the feature vector of the file to be recommended;

[0122] Exemplarily, for the continuous features in the target behavior and the target file, the bucketing method is used for discretization processing to obtain discrete features, and the discrete features are mapped to 64-dimensional embedding vectors, or, for the continuous features in the target behavior and the target file, they are directly mapped to 64-dimensional embedding vectors, and no restrictions are imposed here.

[0123] Among them, the continuous feature is a feature that can take any value. For example, the continuous feature in the target behavior can be the access time of the target object to the target file;

[0124] The discrete feature is a numerical value with a clear category, such as gender, age, etc. Although discrete features can be identified by numbers or characters, the differences between them do not have a quantitative meaning.

[0125] It should be noted that there is no clear sequence between step S103 and step S104. In actual applications, it should be set according to the specific usage scenario, and no restrictions are imposed here.

[0126] S105. Fuse the feature vector of the target file and the time coefficient to obtain the first vector;

[0127] Among them, assuming that the feature vector of the target file is e i , then the first vector is:

[0128] The first vector = β * e i ;

[0129] Among them, the feature vector e of the target file i is a 64-dimensional vector. Since the first vector is the product of the feature vector of the target file and a constant, the scalar multiplication vector is a 64-dimensional vector.

[0130] It can be understood that the description of fusing the feature vector of the target file and the time coefficient here is only an example. In actual applications, the feature vector of the target file and the time coefficient can also be concatenated to obtain the first vector, which is not limited here.

[0131] S106. Input the first vector and the feature vector of the file to be recommended into a neural network to obtain the weight coefficient of the file to be recommended;

[0132] Among them, the neural network is used to generate the weight coefficient of the file to be recommended according to the input feature vectors (the first vector) related to the target file and the feature vector of the file to be recommended.

[0133] Exemplarily, the neural network is an attention neural network as Figure 4 shown.

[0134] Input the first vector and the feature vector of the file to be recommended into the attention neural network. First, process the first vector and the file to be recommended to obtain the outer product of the first vector and the file to be recommended; then concatenate the first vector, the feature vector of the file to be recommended, and the outer product of the feature vector of the file to be recommended and the first vector to obtain a concatenated vector. Input the concatenated vector into an activation function and a linear layer for processing to obtain the weight coefficient of the target file.

[0135] To facilitate a more intuitive understanding of the solution provided by this application, the following introduces the usage method of this attention neural network in a specific scenario:

[0136] When the target object accesses H target files within a preset duration, generating H target behavior records, use method 2) in the aforementioned step S103 to obtain the time coefficient β, and perform an operation similar to the aforementioned step S104 to obtain the first vector. When using the attention neural network as Figure 4 shown, the weight coefficient of the file to be recommended can be calculated using the following formula:

[0137] k = mlp(concat(β * e i ), V i , <β * e i , V i >);

[0138] Among them, k is the weight coefficient of the file to be recommended, and v i is the feature vector of the file to be recommended, and < > represents the outer product operation.

[0139] It can be understood that the introduction of the neural network here is only an example. In actual applications, it should be set according to the specific usage scenario, and no restrictions are imposed here.

[0140] S107. After fusing the weight coefficient of the file to be recommended with the feature vector of the file to be recommended, input it into the fully connected layer to obtain the click-through rate of the file to be recommended;

[0141] After obtaining the weight coefficient of the file to be recommended, fuse the weight coefficient of the file to be recommended with the feature vector of the file to be recommended, and input it into the fully connected layer to obtain the click-through rate of the file to be recommended.

[0142] For example, the weight coefficient of the file to be recommended can be directly concatenated with the feature vector of the file to be recommended, summed, or the corresponding elements between the features are multiplied to achieve the fusion of the weight coefficient of the file to be recommended and the feature vector of the file to be recommended, and the fusion result is input into the fully connected layer, and no restrictions are imposed here.

[0143] Specifically, at least two target feature vectors can be obtained by fusing the weight coefficients of at least two files to be recommended with the feature vectors of at least two files to be recommended respectively; then, after concatenating at least two target feature vectors, input them into the fully connected layer to obtain the click-through rate of the file to be recommended. Among them, before inputting into the fully connected layer, the processing method of the weight coefficients of at least two files to be recommended and the feature vectors of at least two files to be recommended can be as Figure 5a shown.

[0144] Please refer to Figure 5a , Figure 5a for an introduction to the solution with the file to be recommended including the file to be recommended 1, the file to be recommended 2, and the file to be recommended 3; the first vector includes the first vector 1, the first vector 2, and the first vector 3; and the file to be recommended 1 corresponds to the weight coefficient of the file to be recommended 1, the file to be recommended 2 corresponds to the weight coefficient of the file to be recommended 2, and the file to be recommended 3 corresponds to the weight coefficient of the file to be recommended 3; and, taking the first vector 1 corresponding to the file to be recommended 1, the first vector 2 corresponding to the file to be recommended 2, and the first vector 3 corresponding to the file to be recommended 3 as an example.

[0145] After obtaining the weight coefficients of the file to be recommended 1, the weight coefficient of the file to be recommended 2, and the weight coefficient of the file to be recommended 3, the first eigenvector, the second eigenvector, and the third eigenvector are calculated. The first eigenvector is obtained by fusing the weight coefficient of the file to be recommended 1, the first vector 1, and the file to be recommended. The second eigenvector is obtained by fusing the weight coefficient of the file to be recommended 2, the first vector 2, and the file to be recommended. The third eigenvector is obtained by fusing the weight coefficient of the target file to be recommended 3, the first vector 3, and the file to be recommended. The first eigenvector, the second eigenvector, and the third eigenvector are included in at least two target eigenvectors.

[0146] After splicing the first eigenvector, the second eigenvector, and the third eigenvector, they are input into the fully connected layer to obtain the click-through rate of the file to be recommended 1, the click-through rate of the file to be recommended 2, and the click-through rate of the file to be recommended 3. Among them, the click-through rate of the file to be recommended 1, the click-through rate of the file to be recommended 2, and the click-through rate of the file to be recommended 3 are included in the click-through rate of the file to be recommended.

[0147] S108. Recommend the file to be recommended whose click-through rate meets the preset conditions to the target object.

[0148] After obtaining the click-through rate of the file to be recommended, recommend the file to be recommended whose click-through rate meets the preset conditions to the target object. Among them, the click-through rate meeting the preset conditions includes the following situations:

[0149] Situation 1: The preset condition is that the click-through rate of the file to be recommended is greater than the threshold.

[0150] When the click-through rate of the file to be recommended is greater than the threshold, it can be considered that the click-through rate of the file to be recommended meets the preset conditions. Among them, the threshold can be a fixed value preset by the technical personnel.

[0151] Situation 2: Among the N files to be recommended, the files to be recommended with the top 20% click-through rate are the files to be recommended that meet the preset conditions;

[0152] When there are N files to be recommended, sort the click-through rates of the N files to be recommended from large to small, and consider the files to be recommended ranked in the top 5 / N as the files to be recommended that meet the preset conditions.

[0153] It should be noted that the description of the top 20% of the click-through rate here is only an example. In actual applications, the percentage of meeting the preset conditions can be adjusted according to actual needs, and there is no limit here.

[0154] It can be understood that the description of the click-through rate of the file to be recommended meeting the preset conditions here is only an example. In actual applications, it should be set in combination with the specific application scenario, and there is no limit here.

[0155] In the embodiment of the present application, by obtaining a data recommendation request which carries the identifier of the target object, the target behavior, the target file and the file to be recommended are obtained according to the data recommendation request. The target behavior is used to describe the access of the target object to the target file, and the time coefficient is determined according to the target behavior. The time coefficient is used to describe the time series of the target object accessing the target file. Feature extraction is performed on the target behavior, the target file and the file to be recommended to obtain the feature vector of the target behavior, the feature vector of the target file and the feature vector of the file to be recommended. According to the feature vector of the target file and the time coefficient, a first vector is generated. The first vector and the feature vector of the file to be recommended are input into a neural network to obtain the weight coefficient of the file to be recommended. After fusing the weight coefficient of the target file with the feature vector of the file to be recommended, it is input into a fully connected layer to obtain the click-through rate of the file to be recommended. The file to be recommended with a click-through rate meeting the preset condition is recommended to the target object. By calculating the product between the feature vector of the target file and the time coefficient in the Embedding layer of the model, the influence of the time when the target object accesses the target file on the click-through rate prediction is introduced, and without a large amount of time consumption, the influence of time on the prediction result in the user's historical behavior data is enhanced, thereby improving the prediction accuracy.

[0156] In step S103 of the embodiment of the present application, for the second method of determining the time coefficient, when the occurrence moment (timestamp) of the target object accessing the target file is collected, the occurrence moment when the target object accesses the target file can be used as the basis for setting the size of the time coefficient to calculate the time coefficient, avoiding the invalidation of the time coefficient analysis that may be caused when a time series fails, and enhancing the reliability of the solution.

[0157] In step S103 of the embodiment of the present application, for the second formula shown in the second method of determining the time coefficient, after the target behavior is collected, the time coefficient of the target file involved in this target behavior can be calculated by using the sequence of this target behavior in the multiple collected target behaviors. By adopting the non-linear setting of the time coefficient, it is realized that for the time coefficient assigned to the first target behavior before the preset moment, which is closest to the preset moment, is much larger than the time coefficient assigned to the third target behavior before the preset moment. The difference between the time coefficients of different target behaviors is widened, enhancing the flexibility of the solution.

[0158] In step S106 of the embodiment of the present application, Figure 4The neural network structure shown provides a neural network model in which the outer product of the first vector and the file to be recommended is used to realize the fusion of the first vector and the file to be recommended, and the fusion result is concatenated with the first vector and the file to be recommended, and then input into an activation function and a linear layer for processing. The solution provided by this application provides a specific model structure, which, without modifying the model framework, improves the influence of the distance of the occurrence time of the target behavior on the analysis result and improves the accuracy of analyzing the click-through rate of the file to be recommended by the solution.

[0159] In step S107 of the embodiment of this application, a specific implementation manner of analyzing at least two files to be recommended simultaneously is provided, which determines that the solution can process multiple files to be recommended simultaneously to obtain the click-through rate of the files to be recommended, improving the execution efficiency of the solution.

[0160] In step S109 of the embodiment of this application and the modified S108( Figure 5b )), the feature vector of the target object and the scenario feature vector of the target object are introduced into the solution. The feature vector of the target object and the scenario feature vector of the target object are used to assist in analyzing the click-through rate of the file to be recommended, which helps to delimit the interest range of the target object, thereby improving the accuracy of analyzing the click-through rate of the file to be recommended.

[0161] In step S102 of the embodiment of this application Figure 3c , different types of target behaviors are introduced. According to requirements, the target behavior can describe the browsing of the target object for the target file, or the acquisition of the target file by the target object, or the browsing of the target object for the target file and the acquisition of the target file by the target object, expanding the types of target behaviors collected. When analyzing the time coefficient using the target behavior, further settings can be made in combination with different scenarios, improving the flexibility of the solution.

[0162] In some possible implementation manners, the data recommendation method further includes step S109.

[0163] S109. Obtain the feature vector of the target object and the scenario feature vector of the target object.

[0164] Before executing step S107, obtain the feature vector of the target object and the scenario feature vector of the target object.

[0165] Exemplarily, obtain the features of the target object, such as the age and gender of the target object, etc.; obtain the scenario features of the target object, such as the page identifier, time, and the login method of the target object, etc.

[0166] Perform feature mapping on the features of the target object and the scene features of the target object to obtain the feature vector of the target object and the scene feature vector of the target object.

[0167] It should be noted that features such as the age and gender of the target object here are information obtained with the legal authorization of the target object in advance.

[0168] In the updated step S107, first, the weight coefficient of the target file is fused with the feature vector of the file to be recommended to obtain the target feature vector. After concatenating the target feature vector, the feature vector of the target user, and the scene feature vector of the target user, input them into the fully connected layer to obtain the click-through rate of the file to be recommended.

[0169] For details, please refer to Figure 5b , Figure 5b including all the above-mentioned Figure 5a content. On this basis, add the feature vector of the target user and the scene feature vector of the target user. After concatenating the first feature vector, the second feature vector, the third feature vector, the feature vector of the target user, and the scene feature vector of the target user, input them into the fully connected layer to obtain the click-through rate of the file to be recommended 1, the click-through rate of the file to be recommended 2, and the click-through rate of the file to be recommended 3. Among them, the click-through rate of the file to be recommended 1, the click-through rate of the file to be recommended 2, and the click-through rate of the file to be recommended 3 are included in the click-through rate of the file to be recommended.

[0170] In the embodiment of the present application, the feature vector of the target user and the scene feature vector of the target user are obtained, the target feature vector is concatenated with the feature vector of the target user and the scene feature vector of the target user, and input into the fully connected layer to obtain the click-through rate of the file to be recommended. It fully considers various features generated during the execution of the target behavior, improving the integrity of the solution and the accuracy of the prediction result.

[0171] The following will describe in detail the data recommendation device in the present application. Please refer to Figure 6 。 Figure 6 This is a schematic diagram of an embodiment of the data recommendation device 10 in the embodiment of the present application. The data recommendation device 10 includes:

[0172] An acquisition unit 110, configured to acquire a data recommendation request, and the data recommendation request carries the identifier of the target object;

[0173] The acquisition unit 110 is further configured to acquire a target behavior, a target file, and a file to be recommended according to the data recommendation request, where the target behavior is used to describe the access of the target object to the target file;

[0174] A computing unit 120, configured to determine a time coefficient according to a target behavior, where the time coefficient is used to describe the time series of a target object accessing a target file;

[0175] A generating unit 130, configured to perform feature mapping on the target behavior, the target file, and the file to be recommended, to obtain a feature vector of the target behavior, a feature vector of the target file, and a feature vector of the file to be recommended;

[0176] The generating unit 130 is further configured to generate a first vector according to the feature vector of the target file and the time coefficient;

[0177] The computing unit 120 is further configured to input the first vector and the feature vector of the file to be recommended into a neural network to obtain a weight coefficient of the file to be recommended;

[0178] The computing unit 120 is further configured to input the weight coefficient of the file to be recommended and the feature vector of the file to be recommended into a fully connected layer after fusion, to obtain a click-through rate of the file to be recommended;

[0179] A recommending unit 140, configured to recommend the file to be recommended whose click-through rate meets a preset condition to the target object.

[0180] Optionally, the target behavior carries the occurrence time of the target object accessing the target file;

[0181] The computing unit 120 is specifically configured to determine the time coefficient according to the occurrence time of the target object accessing the target file.

[0182] Optionally, the target behavior includes H accesses of the target object to the target file, where H is a positive integer;

[0183] The computing unit 120 is specifically configured to:

[0184] According to the target behavior, determine the proportion of the number of access times when no access occurs at the occurrence time of the first target object accessing the first target file, to obtain a time coefficient to be processed, where the first target object is included in the target object, and the first target file is included in the target file;

[0185] Perform an exponential calculation on the time coefficient to be processed to obtain the time coefficient.

[0186] Optionally, the computing unit 120 is specifically configured to:

[0187] Process the first vector and the feature vector of the file to be recommended to obtain an outer product of the first vector and the feature vector of the file to be recommended;

[0188] Concatenate the first vector, the feature vector of the file to be recommended, and the outer product of the first vector and the feature vector of the file to be recommended to obtain a concatenated vector;

[0189] Input the concatenated vector into an activation function to obtain the weight coefficients of the files to be recommended.

[0190] Optionally, the files to be recommended include at least two files to be recommended;

[0191] The calculation unit 120 is specifically configured to:

[0192] Fuse the weight coefficients of at least two files to be recommended with the feature vectors of the files to be recommended respectively to obtain at least two target feature vectors;

[0193] After concatenating at least two target feature vectors, input them into a fully connected layer to obtain the click-through rate of the files to be recommended.

[0194] Optionally, the acquisition unit 110 is further configured to acquire the feature vector of the target object and the scene feature vector of the target object;

[0195] The calculation unit 120 is specifically configured to concatenate at least two target feature vectors, the feature vector of the target object, and the scene feature vector of the target object, and then input them into a fully connected layer to obtain the click-through rate of the files to be recommended.

[0196] Optionally, the target behavior is used to describe the browsing of the target object for the target file, or the acquisition of the target file by the target object, or the browsing of the target object for the target file and the acquisition of the target file by the target object.

[0197] Figure 7 It is a schematic diagram of a server structure provided by an embodiment of the present application. The server 300 may vary greatly due to configuration or performance differences, and may include one or more central processing units (CPUs) 322 (for example, one or more processors) and a memory 332, and one or more storage media 330 (for example, one or more mass storage devices) for storing application programs 342 or data 344. Among them, the memory 332 and the storage media 330 may be transient storage or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Further, the central processor 322 may be configured to communicate with the storage media 330 and execute a series of instruction operations in the storage media 330 on the server 300.

[0198] The server 300 may further include one or more power supplies 326, one or more wired or wireless network interfaces 350, one or more input / output interfaces 358, and / or one or more operating systems 341, such as Windows Server TM , Mac OS XTM , Unix TM , Linux TM , FreeBSD TM and so on.

[0199] The steps performed by the server in the above embodiments may be based on the Figure 7 server structure shown.

[0200] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0201] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces, and the indirect coupling or communication connection of devices or units can be in electrical, mechanical, or other forms.

[0202] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0203] In addition, the functional units in each embodiment of the present application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0204] When an integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0205] In the embodiments of this application, the term "module" or "unit" refers to a computer program with a predetermined function or a part of a computer program, which works together with other related parts to achieve a predetermined goal, and can be fully or partially implemented by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of an overall module or unit that includes the function of that module or unit.

[0206] The above embodiments are only used to illustrate the technical solutions of this application, rather than to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of various embodiments of this application.

Claims

1. A method for data recommendation, characterized in that, Including: Obtain a data recommendation request, where the data recommendation request carries the identifier of the target object; Obtain a target behavior, a target file, and a file to be recommended according to the data recommendation request, where the target behavior is used to describe the access of the target object to the target file; Determine a time coefficient according to the target behavior, where the time coefficient is used to describe the time series of the target object accessing the target file; Perform feature mapping on the target behavior, the target file, and the file to be recommended to obtain a feature vector of the target behavior, a feature vector of the target file, and a feature vector of the file to be recommended; Generate a first vector according to the feature vector of the target file and the time coefficient; Input the first vector and the feature vector of the file to be recommended into a neural network to obtain a weight coefficient of the file to be recommended; After fusing the weight coefficient of the file to be recommended and the feature vector of the file to be recommended, input it into a fully connected layer to obtain the click-through rate of the file to be recommended; Recommend the file to be recommended whose click-through rate meets a preset condition to the target object.

2. The method according to claim 1, characterized in that, The target behavior carries the access occurrence time of the target object to the target file; The determining the time coefficient according to the target behavior includes: Determine the time coefficient according to the access occurrence time of the target object to the target file.

3. The method according to claim 2, wherein The target behavior includes H times of the target object's access to the target file, where H is a positive integer; The determining the time coefficient according to the access occurrence time of the target object to the target file includes: According to the target behavior, determine the proportion of the number of unaccessed accesses at the access occurrence time of the target object to the first target file, and obtain the time coefficient to be processed, where the first target file is included in the target file; Perform an exponential calculation on the time coefficient to be processed to obtain the time coefficient.

4. The method according to any one of claims 1 to 3, characterized in that The inputting the first vector and the feature vector of the file to be recommended into a neural network to obtain the weight coefficient of the file to be recommended includes: Process the first vector and the feature vector of the file to be recommended to obtain the outer product of the first vector and the feature vector of the file to be recommended; Concatenate the first vector, the feature vector of the file to be recommended, and the outer product of the first vector and the feature vector of the file to be recommended to obtain a concatenated vector; Input the concatenated vector into an activation function to obtain the weight coefficient of the file to be recommended.

5. The method according to claim 4, characterized in that, The file to be recommended includes at least two files to be recommended; The inputting the weight coefficient of the file to be recommended and the feature vector of the file to be recommended into a fully connected layer to obtain the click-through rate of the file to be recommended includes: Fuse the weight coefficients of the at least two files to be recommended and the feature vectors of the at least two files to be recommended respectively to obtain at least two target feature vectors; After concatenating the at least two target feature vectors, input them into a fully connected layer to obtain the click-through rate of the file to be recommended.

6. The method according to claim 5, wherein The method further includes: Obtain the feature vector of the target object and the scene feature vector of the target object; After concatenating the at least two target feature vectors and inputting them into a fully-connected layer, obtaining the click-through rate of the file to be recommended includes: After concatenating the at least two target feature vectors, the feature vector of the target object, and the scene feature vector of the target object, inputting them into a fully-connected layer, obtaining the click-through rate of the file to be recommended.

7. The method according to any one of claims 1 to 3, characterized in that The target behavior is used to describe the browsing of the target object for the target file, or the acquisition of the target file by the target object, or the browsing of the target object for the target file and the acquisition of the target file by the target object.

8. A device for data recommendation, characterized in that, It includes: An acquisition unit, configured to acquire a data recommendation request, where the data recommendation request carries an identifier of a target object; The acquisition unit is further configured to acquire a target behavior, a target file, and a file to be recommended according to the data recommendation request, where the target behavior is used to describe the access of the target object to the target file; A calculation unit, configured to determine a time coefficient according to the target behavior, where the time coefficient is used to describe the time series of the target object accessing the target file; A generation unit, configured to perform feature mapping on the target behavior, the target file, and the file to be recommended, obtaining a feature vector of the target behavior, a feature vector of the target file, and a feature vector of the file to be recommended; The generation unit is further configured to generate a first vector according to the feature vector of the target file and the time coefficient; The calculation unit is further configured to input the first vector and the feature vector of the file to be recommended into a neural network, obtaining a weight coefficient of the file to be recommended; The calculation unit is further configured to fuse the weight coefficient of the file to be recommended and the feature vector of the file to be recommended and input them into a fully-connected layer, obtaining the click-through rate of the file to be recommended; A recommendation unit, configured to recommend the file to be recommended whose click-through rate meets a preset condition to the target object.

9. A computer device, characterized in that, It includes: A memory, a transceiver, a processor, and a bus system; Wherein, the memory is used to store a program; The processor is configured to execute the program in the memory, including executing the data recommendation method according to any one of claims 1 to 7; The bus system is used to connect the memory and the processor, enabling the memory and the processor to communicate.

10. A computer-readable storage medium, including instructions, which when running on a computer, cause the computer to execute the data recommendation method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, The computer program is executed by a processor to perform the data recommendation method according to any one of claims 1 to 7.

Images