Content recommendation method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a content recommendation method and device, electronic equipment and a storage medium, and aims to improve the content recommendation accuracy. The method comprises the following steps: on the basis of the correlation between each browsing duration feature of a target object and a historical behavior feature, respectively weighting historical recommendation content corresponding to each browsing duration feature and a feature element corresponding to the historical behavior feature to obtain a corresponding object interest feature; performing feature fusion based on the object attribute feature of the target object, the content attribute feature of the to-be-recommended content and the interest feature of each object to obtain a target fusion feature; and pushing the to-be-recommended content to the target object when an estimated value obtained based on the target fusion feature meets a preset recommendation condition. Due to the fact that the interest preferences of different levels of the target object can be represented through the obtained interest features of all the objects, the prediction value of the to-be-recommended content is more accurate, and the content recommendation accuracy is improved.

Description

Background Art

[0002] With the development of Internet technology, an object can obtain a variety of multimedia content through a network platform. When an object obtains multimedia content, the network platform can deliver multimedia content that the object may be interested in according to the object's interest preferences.

[0003] In related technologies, the interest preferences of an object are usually estimated based on the click behavior of the object on the delivered multimedia content.

[0004] However, an object's interests are diverse. Even if an object clicks on a certain multimedia content, it does not necessarily mean that the object is interested in this multimedia content. Therefore, only based on the object's click behavior, the interest preferences of the object cannot be effectively captured, resulting in a relatively low accuracy of the recommended multimedia content. Summary of the Invention

[0005] Embodiments of the present application provide a content recommendation method, apparatus, electronic device, and storage medium to improve the accuracy of content recommendation.

[0006] A content recommendation method provided by an embodiment of the present application includes:

[0007] Based on the correlation between each browsing duration feature of a target object and a historical behavior feature, respectively weight the feature elements corresponding to the historical behavior feature for each historical recommended content corresponding to each browsing duration feature to obtain corresponding object interest features; the historical behavior feature is generated based on each historical recommended content clicked by the target object, and each browsing duration feature is generated based on the browsing duration of the corresponding historical recommended content;

[0008] Based on the object attribute feature of the target object, the content attribute feature of the content to be recommended, and the obtained object interest features, perform feature fusion to obtain a target fusion feature;

[0009] Based on the target fusion feature, perform prediction to obtain an estimated value of the content to be recommended;

[0010] When the estimated value meets a preset recommendation condition, push the content to be recommended to the target object.

[0011] A content recommendation apparatus provided by an embodiment of the present application includes:

[0012] A processing unit, configured to, based on the correlation between each browsing duration feature of a target object and a historical behavior feature, respectively weight the feature elements corresponding to the historical recommended content corresponding to each browsing duration feature in the historical behavior feature, to obtain corresponding object interest features; the historical behavior feature is generated based on each historical recommended content clicked by the target object, and each browsing duration feature is generated based on the browsing duration of the corresponding historical recommended content;

[0013] A fusion unit, configured to perform feature fusion based on the object attribute feature of the target object, the content attribute feature of the content to be recommended, and the obtained object interest features, to obtain a target fusion feature;

[0014] A prediction unit, configured to perform prediction based on the target fusion feature, to obtain an estimated value of the content to be recommended;

[0015] A push unit, configured to, when the estimated value meets a preset recommendation condition, push the content to be recommended to the target object.

[0016] Optionally, the processing unit is specifically configured to:

[0017] For each of the browsing duration features, respectively perform the following operations:

[0018] Based on the correlation between one browsing duration feature and the historical behavior feature, obtain a weight feature;

[0019] Based on the weight feature, weight the feature elements corresponding to the historical recommended content corresponding to one browsing duration feature in the historical behavior feature, to obtain a corresponding object interest feature.

[0020] Optionally, the processing unit is specifically configured to:

[0021] Perform feature crossing on one browsing duration feature and the historical behavior feature, to obtain a behavior crossing feature;

[0022] Perform downsampling on the behavior crossing feature, to obtain a sampled feature;

[0023] Perform normalization processing on the sampled feature, to obtain the weight feature.

[0024] Optionally, the apparatus further includes a partitioning unit, configured to obtain each of the browsing duration features in the following manner:

[0025] Based on at least one endpoint value, partition the browsing duration corresponding to each of the historical recommended contents, to obtain corresponding duration sets;

[0026] For the obtained multiple duration sets, respectively perform the following operations:

[0027] For a set of durations, among the viewing durations corresponding to the respective historical recommended contents, the viewing durations other than those included in the set of durations are used as candidate viewing durations, and the obtained candidate viewing durations are updated to a preset value.

[0028] Based on the viewing durations included in the set of durations and the updated candidate viewing durations, corresponding viewing duration features are obtained.

[0029] Optionally, the at least one endpoint value includes a first endpoint value and a second endpoint value, and the multiple sets of durations include a first set of durations, a second set of durations, and a third set of durations.

[0030] Then the partitioning unit is specifically configured to:

[0031] For the respective viewing durations, the viewing durations less than the first endpoint value are partitioned into the first set of durations, the viewing durations greater than the second endpoint value are partitioned into the second set of durations, and the viewing durations not less than the first endpoint value and not greater than the second endpoint value are partitioned into the third set of durations.

[0032] Optionally, among the first ratio and the second ratio obtained based on the respective viewing durations, the first ratio is less than the second ratio; wherein, the first ratio is the ratio of the number of viewing durations less than the first endpoint value to the number of all viewing durations among the respective viewing durations, and the second ratio is the ratio of the number of viewing durations less than the second endpoint value to the number of all viewing durations among the respective viewing durations.

[0033] Optionally, the fusion unit is specifically configured to:

[0034] Based on the object attribute feature, the content attribute feature, and the respective object interest features, perform multiple rounds of feature cross - ing to obtain multiple information cross - ing features, where the two features selected in different rounds are not completely the same.

[0035] Based on the multiple information cross - ing features, perform feature fusion to obtain the target fusion feature.

[0036] Optionally, the fusion unit is specifically configured to:

[0037] In one round of feature cross - ing, select two different features from the object attribute feature, the content attribute feature, and the respective object interest features.

[0038] Based on the inner - product result of the two selected features, obtain the corresponding information cross - ing feature.

[0039] Optionally, the fusion unit is specifically configured to:

[0040] Based on each first preset weight, perform weighted fusion on the multiple information cross - features to obtain a first fusion feature;

[0041] Based on each second preset weight, perform weighted fusion on the object attribute feature, the content attribute feature, and each object interest feature to obtain a second fusion feature;

[0042] Perform feature fusion on the first fusion feature, the second fusion feature, and a preset feature to obtain the target fusion feature.

[0043] Optionally, the fusion unit is specifically configured to:

[0044] Based on each third preset weight, perform weighted fusion on the object attribute feature, the content attribute feature, and each object interest feature to obtain the target fusion feature.

[0045] An electronic device provided by an embodiment of the present application includes a processor and a memory. Among them, the memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of any one of the above - mentioned content recommendation methods.

[0046] An embodiment of the present application provides a computer - readable storage medium, which includes a computer program. When the computer program runs on an electronic device, the computer program is used to cause the electronic device to execute the steps of any one of the above - mentioned content recommendation methods.

[0047] An embodiment of the present application provides a computer program product. The computer program product includes a computer program, and the computer program is stored in a computer - readable storage medium; when a processor of an electronic device reads the computer program from the computer - readable storage medium, the processor executes the computer program, causing the electronic device to execute the steps of any one of the above - mentioned content recommendation methods.

[0048] The beneficial effects of the present application are as follows:

[0049] The embodiments of the present application provide a content recommendation method, apparatus, electronic device, and storage medium. During the content browsing process of a target object, based on the historical recommended contents clicked to generate historical behavior features, multiple browsing duration features are generated based on the browsing durations of the corresponding historical recommended contents. Then, based on the correlations between the respective browsing duration features and the historical behavior features, the corresponding feature elements in the historical behavior features are weighted respectively to obtain the corresponding object interest features. In this way, the browsing duration of the historical recommended contents is introduced as supplementary information for capturing the object's interest, and the preference degrees of the target object for different historical recommended contents are mined through the browsing duration. The generated object interest features can fully represent the different levels of interest preferences of the target object for the content.

[0050] Furthermore, based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the obtained object interest features, feature fusion is performed to obtain the target fusion feature. The obtained target fusion feature can more accurately express the relationship between the interest preference of the target object and the content to be recommended. Therefore, when making a prediction based on the target fusion feature, the obtained predicted value is more accurate. When the predicted value meets the preset recommendation condition, the content to be recommended is pushed to the target object, thus improving the accuracy of content recommendation.

[0051] Other features and advantages of the present application will be described in the subsequent specification, and some of them will become obvious from the specification or be understood by implementing the present application. The objectives and other advantages of the present application can be achieved and obtained through the structures specifically pointed out in the written specification, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0053] Figure 1 is an optional schematic diagram of an application scenario in the embodiments of the present application;

[0054] Figure 2 is the implementation flowchart of a content recommendation method in the embodiments of the present application;

[0055] Figure 3 is a schematic diagram of a browsing duration division method in the embodiments of the present application;

[0056] Figure 4 is a schematic diagram of the acquisition process of browsing duration features in the embodiments of the present application;

[0057] Figure 5Schematic diagram of the acquisition process of object interest features in the embodiments of the present application;

[0058] Figure 6 Schematic diagram of a subscription number message page in the embodiments of the present application;

[0059] Figure 7 Schematic diagram of a feature fusion method in the embodiments of the present application;

[0060] Figure 8 Schematic diagram of a recommendation condition in the embodiments of the present application;

[0061] Figure 9 Schematic diagram of a feature weighting method in the embodiments of the present application;

[0062] Figure 10 Schematic diagram of a feature crossing process in the embodiments of the present application;

[0063] Figure 11 Schematic diagram of the structure of a ranking model in the embodiments of the present application;

[0064] Figure 12 Schematic diagram of the logic of a content recommendation method in the embodiments of the present application;

[0065] Figure 13 Schematic diagram of the structure of a content recommendation device in the embodiments of the present application;

[0066] Figure 14 Schematic diagram of a hardware composition structure of an electronic device applying the embodiments of the present application;

[0067] Figure 15 Schematic diagram of a hardware composition structure of another electronic device applying the embodiments of the present application. Detailed implementation manners

[0068] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the technical solutions of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments recorded in this application document without creative efforts shall fall within the scope of protection of the technical solutions of the present application.

[0069] Some concepts involved in the embodiments of the present application are introduced below.

[0070] Estimated value: It refers to the feedback of the target object on the content to be recommended predicted through the target fusion features, which may include information such as click-through rate (CTR), conversion rate (CVR), and conversion rate after click.

[0071] Click-through rate: It is defined as the ratio of the clicked delivery information to the delivery information in the effective exposure stage, and is used to characterize the probability of the delivery information being clicked from the effective exposure stage.

[0072] Label: The labeled data relied on when training a deep neural network. For example, "0 / 1" represents "belonging to / not belonging to" a certain category.

[0073] Embedding: A numerical vector composed of multiple floating-point numbers, which describes various attributes and properties of an item or a user in a high-dimensional space. In the embodiments of the present application, features such as browsing duration feature, historical behavior feature, object attribute feature, and content attribute feature can be obtained through Embedding. Attention model: A machine learning model, usually used to process sequence data, such as text data in natural language processing. Among them, the Query-Key-Value model is an implementation method of the Attention model. By mapping the input sequence into Query, Key, and Value vectors respectively, then calculating the similarity between Query and Key to obtain the Attention weight, and finally using the Attention weight to perform weighted summation on the Value vector to obtain the output vector. For different models, the specific definitions and design forms of specific Q, K, and V are different, and the methods of calculating similarity and weighted summation will also vary.

[0074] The Query-Key-Value model can be applied to various tasks, such as machine translation, natural language generation, etc., and is also widely used in recommendation ranking models. The advantage is that it can perform weighted processing on information at different moments in the sequence, so as to better capture the relationships in the sequence and improve the performance of the model. In the embodiments of the present application, the historical behavior features can be weighted through the Query-Key-Value model, that is, using the browsing duration feature as the Key, and the historical behavior features as the Query and Value.

[0075] The embodiments of the present application relate to artificial intelligence (AI) and machine learning technologies, and are designed based on natural language processing technology and machine learning (ML) in artificial intelligence.

[0076] Artificial Intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best 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 way similar 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.

[0077] Artificial intelligence technology is an interdisciplinary subject with a wide range of fields involved, 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, autonomous driving, and intelligent transportation.

[0078] Machine learning is an interdisciplinary subject that involves multiple fields 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.

[0079] Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence, and its applications cover all fields of artificial intelligence. Machine learning and deep learning generally include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, and inductive learning. An artificial neural network (ANN) abstracts the human brain neuron network from the perspective of information processing, establishes a certain simple model, and forms different networks according to different connection methods. A neural network is an operation model composed of a large number of nodes (or neurons) connected to each other. Each node represents a specific output function, called an activation function. The connection between each two nodes represents a weighted value for the signal passing through this connection, which is called a weight. This is equivalent to the memory of the artificial neural network. The output of the network varies depending on the connection method of the network, the weight value, and the activation function. Usually, the network itself is an approximation of a certain algorithm or function in nature, or may be an expression of a logical strategy.

[0080] The solution provided in the embodiments of this application relates to the machine learning technology of artificial intelligence. The content recommendation method in the embodiments of this application can be implemented through a neural network model. Specifically, a content recommendation model is obtained by using machine learning technology with sample recommended content, and then, in the actual content recommendation process, the learned content recommendation model is used to evaluate the content to be recommended to obtain a corresponding predicted value.

[0081] Specifically, the embodiments of this application involve a training part and an application part. Among them, in the training part, the content recommendation model is trained through the technology of machine learning, and the model parameters are continuously adjusted through an optimization algorithm until the model converges; the application part is used to evaluate the parameter values using the content recommendation model trained in the training part, and then conduct content recommendation. In addition, it should be noted that the evaluation in the embodiments of this application can be online training or offline training, which is not specifically limited here. In this article, offline training is used as an example for illustration.

[0082] The design concept of the embodiments of this application is briefly introduced below:

[0083] With the development of Internet technology, an object can obtain a variety of multimedia content through a network platform. When the object obtains multimedia content, the network platform can deliver multimedia content that the object may be interested in according to the object's interest preferences.

[0084] Accurate user modeling is crucial for online personalized recommendation services. Generally speaking, users have diverse interests and are not limited to a single aspect. For example, when browsing the content pushed in the information stream, users may show interests in animals, dance, and food; users may also have different interests in short and fast-paced entertainment content and in-depth analysis articles.

[0085] In related technologies, generally, the interest preferences of an object are estimated based on the click behavior of the object on the multimedia content that has been delivered. However, an object's interests are diverse, and even if an object clicks on a certain multimedia content, it does not necessarily mean that the object is interested in this multimedia content.

[0086] Furthermore, conventional ranking models tend to encode a user's historical behavior into a single representation vector. For example, the sequence of the identity document (ID) numbers of the items that the user has historically clicked is encoded into a single representation vector. However, the method of encoding into a single representation vector does not have sufficient ability to effectively capture the different interests of the user. Thus, related research methods focus on studying how to extract multiple and different interest representation vectors from the user's historical behavior, specifically including the following methods:

[0087] (1) Based on a dynamic routing mechanism (using a capsule network as a computing module), adaptively aggregate multiple representation vectors of the user from the user's historical behavior, believing that these multiple vectors can express different aspects of the user's interests;

[0088] (2) Use a multi-head self-attention network to extract multiple representation vectors of the user;

[0089] (3) Introduce diversity measurement and aggregate to generate multiple user representation vectors with diversity;

[0090] (4) Aggregate different user representations through a graph neural network.

[0091] The above methods all obtain multiple aggregation results for the user's historical behavior through modifications to the model structure or calculation methods. Due to the lack of introduction of prior information that explicitly describes the different interests of the user, the main concern for these models is whether the multiple user representation vectors extracted can accurately represent the user's differentiated interest preferences.

[0092] There are also methods that propose to additionally introduce the category attributes of the messages as supplementary information and design a category-aware Attention network to incorporate the category attributes as explicit interest signals into the modeling. The multi-interest modeling based on category attributes as supplementary information has rationality, but it cannot capture the deep and shallow interest preferences of the user.

[0093] Therefore, based on the above methods, the interest preferences of the object are effectively captured, resulting in a relatively low accuracy of the recommended multimedia content.

[0094] In view of this, the embodiments of the present application provide a content recommendation method, device, electronic device, and storage medium. During the content browsing process of the target object, based on the historical recommended content clicked, historical behavior features are generated. Based on the browsing duration of the corresponding historical recommended content, multiple browsing duration features are generated. Then, based on the correlation between each browsing duration feature and the historical behavior features, the corresponding feature elements in the historical behavior features are weighted respectively to obtain the corresponding object interest features. In this way, the browsing duration of the historical recommended content is introduced as supplementary information for capturing the object's interest. By mining the preference degree of the target object for different historical recommended content through the browsing duration, the generated object interest features can fully represent the different levels of interest preferences of the target object for the content.

[0095] Furthermore, based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the obtained object interest features, feature fusion is performed to obtain the target fusion feature. The obtained target fusion feature can more accurately express the relationship between the interest preferences of the target object and the content to be recommended. Therefore, when making a prediction based on the target fusion feature, the obtained predicted value is more accurate. When the predicted value meets the preset recommendation conditions, the content to be recommended is pushed to the target object, achieving an improvement in the accuracy of content recommendation.

[0096] The following describes the preferred embodiments of the present application with reference to the accompanying drawings of the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. And without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0097] As Figure 1 shown, it is a schematic diagram of the application scenario of the embodiments of the present application. The application scenario diagram includes two terminal devices 110 and a server 120.

[0098] In the embodiments of the present application, the terminal device includes, but is not limited to, devices such as mobile phones, tablet computers, laptop computers, desktop computers, e-book readers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, etc.; a client related to content recommendation may be installed on the terminal device, and the client may be software (such as a browser, a video application, a shopping application, an instant messaging application, etc.), or a web page, a small program, etc., and the server is the background server corresponding to the software or the web page, the small program, etc., or a server dedicated to content recommendation, and the present application does not make specific limitations. The server may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing 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 (Content Delivery Network, CDN), and big data and artificial intelligence platforms.

[0099] It should be noted that the content recommendation method in the embodiments of the present application may be executed by an electronic device, and the electronic device may be a server or a terminal device, that is, the method may be executed independently by the server or the terminal device, or jointly by the server and the terminal device. For example, when jointly executed by the server and the terminal device, the server weights the corresponding feature elements in the historical behavior features for the historical recommended content corresponding to each browsing duration feature respectively based on the correlation between each browsing duration feature of the target object and the historical behavior features, obtains the corresponding object interest features, and performs feature fusion based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the obtained object interest features to obtain the target fusion features, and finally makes a prediction based on the target fusion features to obtain the predicted value of the content to be recommended. The server sends the content to be recommended and the predicted value to the terminal device, and when the predicted value meets the preset recommendation conditions, the terminal device pushes the content to be recommended to the target object.

[0100] In an alternative embodiment, the terminal device and the server may communicate through a communication network.

[0101] In an alternative embodiment, the communication network is a wired network or a wireless network.

[0102] It should be noted that Figure 1 The above is only an example, and actually the number of terminal devices and servers is not limited, and no specific limitations are made in the embodiments of the present application.

[0103] In the embodiments of the present application, when the number of servers is multiple, the multiple servers can form a blockchain, and the servers are nodes on the blockchain; as for the content recommendation method disclosed in the embodiments of the present application, the content to be recommended, historical recommended content, etc. involved therein can all be stored on the blockchain.

[0104] In addition, the embodiments of the present application can be applied to various scenarios, including not only content recommendation scenarios, but also, without limitation, scenarios such as cloud technology, artificial intelligence, intelligent transportation, assisted driving, games, etc.

[0105] Next, in combination with the application scenarios described above, the content recommendation method provided by the exemplary embodiment of the present application will be described with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown for the convenience of understanding the spirit and principle of the present application, and the embodiments of the present application are not limited in this regard.

[0106] As Figure 2 shown, it is a flowchart of the implementation of a content recommendation method provided by an embodiment of the present application. Taking the execution entity as a server as an example, the specific implementation process of this method includes the following steps S21 - S24:

[0107] S21: The server, based on the correlation between each browsing duration feature of the target object and the historical behavior feature respectively, weights the feature elements corresponding to the historical recommended content corresponding to each browsing duration feature in the historical behavior feature, and obtains the corresponding object interest feature;

[0108] Specifically, the historical recommended content is the multimedia content that has been pushed to the target object. Correspondingly, the target object can perform some operations on the historical object content, such as click, forward, comment and other operations. The historical behavior feature in the embodiments of the present application is mainly generated based on each historical recommended content clicked by the target object. By analyzing the clicked historical recommended content, it can help to dig out the content that the target object is interested in.

[0109] Furthermore, since the historical recommended content clicked by the target object is not necessarily the content that the target object is interested in. Generally speaking, for the content that the user (target object) has a deep preference for, the user will spend a longer time browsing it thoroughly. On the contrary, for the content with a lower degree of interest, the reading duration after clicking is relatively short. Therefore, the embodiments of the present application propose to use the reading duration sequence as supplementary information to accurately and efficiently model the user's different levels of interest preferences.

[0110] Specifically, use: 1) The ID sequence of the items (historical recommended content) that the user has clicked within a past period of time (such as 30 days) (also called the user's historical behavior sequence); 2) Corresponding to the historical behavior sequence, the reading duration (browsing duration) of each clicked item, which is also a sequence. It can be seen that 2) is the supplementary information of 1).

[0111] For example, the historical behavior sequence is [1001, 1003, 1004, 1007, 1010], where each element in the sequence is the ID of the historical recommended content. "1001" is the ID of the historical recommended content 1. The corresponding browsing duration sequence for the historical behavior sequence is [0.1, 1, 3, 4, 0.5], and each element in the sequence is the browsing duration of the corresponding historical recommended content. For example, the browsing duration of the historical recommended content 1 is 0.1 minute.

[0112] After mapping the historical behavior sequence and the browsing duration sequence through the Embedding layer, historical behavior features and browsing duration features can be obtained. For a given browsing duration sequence, assuming its length is 256, denoted as s1, then s1 is a vector in R^{256X1}. Correspondingly, for the historical behavior sequence, after being mapped through the Embedding layer, it is s2, then s2 ∈ R^{256XD}, where D is the embedding size.

[0113] It should be noted that the features involved in the embodiments of the present application can all be in the form of vectors, and will not be listed one by one here.

[0114] Furthermore, the entire browsing duration sequence can be directly mapped to a browsing duration feature, or the browsing duration sequence can be divided, and corresponding browsing duration features are generated according to the browsing durations of the historical recommended content divided into the same set. That is, each browsing duration feature is generated based on the browsing duration of the corresponding historical recommended content.

[0115] When dividing the browsing duration, it can be divided according to the chronological order of the push times of the corresponding historical recommended content. For example, those with a push time earlier than 1 year are divided into one set, those with a push time between 1 year and 3 months are divided into one set, and those with a push time within 3 months are divided into one set.

[0116] It can also be divided according to the level of the browsing duration, such as Figure 3 As shown, it is a schematic diagram of a method for dividing the browsing duration in the embodiments of the present application. Those with a browsing duration greater than 5 minutes are divided into set 1, those with a browsing duration between 1 minute and 5 minutes are divided into set 2, and those with a browsing duration less than 1 minute are divided into set 3.

[0117] It should be noted that the above is only an example for explaining the rules for dividing the browsing duration. In fact, it can also be divided based on other methods, and no specific limitation is made here.

[0118] Optionally, the browsing duration is divided in the following way: based on at least one endpoint value, the browsing durations corresponding to each historical recommended content are divided to obtain corresponding duration sets.

[0119] When the number of endpoint values is 1, two duration sets can be obtained through partitioning. When the number of endpoint values is greater than 1, the number of duration sets obtained through partitioning is the number of endpoint values plus one.

[0120] For example, expressing the browsing duration in minutes, the browsing durations corresponding to each historical recommended content include: 0.1, 2, 6, 1, 0.3, 0.8, 4, 5, 3.1, 4, 0.7. The first endpoint value is 1, and the second endpoint value is 3. The obtained duration set 1 {0.1, 0.3, 0.8, 0.7}, duration set 2 {2, 1}, and duration set 3 {6, 4, 5, 3.1, 4}.

[0121] Taking the endpoint value of 2 as an example, the obtained duration set 4 {0.1, 1, 0.3, 0.8, 0.7} and duration set 5 {2, 6, 4, 5, 3.1, 4}.

[0122] Taking the first endpoint value and the second endpoint value as examples, the browsing durations are partitioned in the following way:

[0123] For each browsing duration, the browsing durations less than the first endpoint value are partitioned into the first duration set, the browsing durations greater than the second endpoint value are partitioned into the second duration set, and the browsing durations not less than the first endpoint value and not greater than the second endpoint value are partitioned into the third duration set.

[0124] For example, each browsing duration includes: 2.5, 0.2, 0.5, 0.7, 3, 5, 4.3, 0.5, 0.1, 4, 0.7. The first endpoint value is 2, and the second endpoint value is 3. The obtained first duration set {0.2, 0.5, 0.7, 0.5, 0.1, 0.7}, second duration set {2.5}, and third duration set {3, 5, 4.3, 4}.

[0125] The first endpoint value and the second endpoint value can be set according to requirements. Optionally, the following relationship exists between the first endpoint value and the second endpoint value:

[0126] Among the first ratio and the second ratio obtained based on each browsing duration, the first ratio is less than the second ratio; wherein, the first ratio is: the ratio of the number of browsing durations less than the first endpoint value to the number of all browsing durations, and the second ratio is: the ratio of the number of browsing durations less than the second endpoint value to the number of all browsing durations.

[0127] Specifically, when determining the first endpoint value and the second endpoint value, the first ratio and the second ratio are introduced. For example, if the first ratio is 0.3 and the second ratio is 0.4, then the first endpoint value is the 30% quantile of all browsing durations, and the second endpoint value is the 40% quantile of all browsing durations.

[0128] For a browsing duration sequence [2.5, 0.2, 0.5, 0.7, 3, 4.3, 0.5, 0.1, 4, 0.7], after arranging the included browsing durations in ascending order, it becomes [0.1, 0.2, 0.5, 0.5, 0.7, 0.7, 2.5, 3, 4, 4.3]. The first ratio is 0.2 and the second ratio is 0.7. That is, the first endpoint value is the 20% quantile of the browsing durations, and the second endpoint value is the 70% quantile of the browsing durations. Then the first endpoint value can be taken as 0.3. The ratio of the number of browsing durations less than the first endpoint value to the number of all browsing durations is 0.3. The second endpoint value is taken as 2.6. The ratio of the number of browsing durations less than the second endpoint value to the number of all browsing durations is 0.7, which meets the condition that the first ratio is less than the second ratio.

[0129] Still taking the browsing durations arranged in ascending order as [0.1, 0.2, 0.5, 0.5, 0.7, 0.7, 2.5, 3, 4, 4.3], with the first ratio being 0.2 and the second ratio being 0.7 as an example, it can also be that the first endpoint value is 0.4 and the second endpoint value is 2.7, or the first endpoint value is 0.4 and the second endpoint value is 2.8. Here, they are not listed one by one.

[0130] The magnitudes of the first ratio and the second ratio can be set according to actual needs. In the embodiments of this application, mainly taking the first ratio as 0.2 and the second ratio as 0.7 as an example for illustration.

[0131] The above introduces the method of dividing the browsing duration to obtain the duration set in the embodiments of this application. After obtaining multiple duration sets, for one duration set, perform the following operations:

[0132] Among the browsing durations corresponding to each historical recommended content, the browsing durations other than those included in one duration set are used as candidate browsing durations, and the obtained candidate browsing durations are updated to preset values; based on the browsing durations included in one duration set and the updated candidate browsing durations, the corresponding browsing duration features are obtained.

[0133] As Figure 4 shown, it is a schematic diagram of the acquisition process of the browsing duration features in the embodiments of this application. Taking the browsing durations as [2.5, 0.2, 0.5, 0.7, 3, 4.3, 0.5, 0.1, 4, 0.7] as an example, by dividing the browsing durations, duration set a {0.2, 0.1}, duration set b {2.5, 0.5, 0.7, 0.5, 0.7}, and duration set c {3, 4.3, 4} are obtained.

[0134] For the duration set a {0.2, 0.1}, the candidate browsing durations are 2.5, 0.5, 0.7, 3, 4.3, 0.5, 4, 0.7. The preset value can be set according to requirements. Taking the preset value as 0 as an example, each candidate browsing duration is updated to 0. The browsing duration sequence 1 composed of the browsing durations included in the duration set a and each updated candidate browsing duration is [0, 0.2, 0, 0, 0, 0, 0, 0.1, 0, 0]. The browsing duration sequence is mapped through the Embedding layer to obtain the browsing duration feature 1.

[0135] The duration set b {2.5, 0.5, 0.7, 0.5, 0.7}, and the corresponding browsing duration sequence 2 is [2.5, 0, 0.5, 0.7, 0, 0, 0.5, 0, 0, 0.7]. The browsing duration sequence 2 is mapped through the Embedding layer to obtain the browsing duration feature 2.

[0136] The duration set c {3, 4.3, 4}, and the corresponding browsing duration sequence 3 is [0, 0, 0, 0, 3, 4.3, 0, 0, 4, 0]. The browsing duration sequence 3 is mapped through the Embedding layer to obtain the browsing duration feature 3.

[0137] The embodiment of the present application proposes a multi-key index self-attention module (Multi-Key Attention Module) based on the browsing duration sequence, which is used to weight the historical behavior features in step S21.

[0138] As Figure 5 shown, it is a schematic diagram of the acquisition process of the object interest feature in the embodiment of the present application. For a given browsing duration sequence, assuming its length is 256, denoted as s1, then s1 is a vector in R^{256×1}. Correspondingly, the clicked item ID sequence (historical behavior sequence), after being mapped through the Embedding layer, is s2 (historical behavior feature), then s2 ∈ R^{256×D}, where D is the embedding size.

[0139] Stat module: For each reading duration sequence (browsing duration sequence), calculate its 20% and 70% quantiles (denoted as r1 and r2) to obtain each user's personalized reading duration habit.

[0140] Filter: According to the calculated r1 and r2, divide the reading duration sequence into "low duration sequence", "medium duration sequence" and "high duration sequence". Taking the "low duration sequence" as an example, only the numbers in the sequence less than r1 are retained, and the other numbers are set to 0. Similarly, the "medium duration sequence" retains the numbers with values between r1 and r2, and the "high duration sequence" retains the numbers higher than r2.

[0141] Within the Multi-Key Attention Module, three self-attention modules (Attention-module) are used to model low-, medium-, and high-duration sequences respectively, extracting three different levels of user interest representation vectors (object interest features). Specifically as follows:

[0142] For the low-duration sequence l1, it first passes through a linear transformation layer (Linear), mapped to f(l1) ∈ R^{256×1}, serving as the Key value of the Attention module. The embedded click item ID sequence is used as the Query and Value values. Query and Key are multiplied point by point for corresponding items, and then average pooling (meanpooling) and normalization (SoftMax) operations are performed on the dimension of the embedding size (obtaining a weight vector with the sum of elements equal to 1) to get the attention weight A:

[0143] A″ = Q ⊙ K ∈ R 256×D

[0144] A′ = MeanPool(A″) ∈ R 256×1

[0145] A = SoftMax(A′) ∈ R 256×1

[0146] The weight vector A is used to perform matrix multiplication on Value to achieve weighted aggregation of each item Embedding in the sequence, obtaining the user's shallow interest representation vector U1:

[0147] U1 = A T V ∈ R 1×D

[0148] Similarly, the Attention-module models with the medium-duration sequence l2 as the input Key value, using the embedded click item ID sequence as the Query and Value, obtaining the user's regular interest representation vector U2; models with the high-duration sequence l3 as the input Key value, using the embedded click item ID sequence as the Query and Value, obtaining the user's deep interest representation vector U3.

[0149] The proposed multi-key index self-attention module based on the reading duration sequence above can be conveniently embedded into the mainstream recommendation ranking model, providing multi-level user interest representations as additional features.

[0150] Step S21 mines the interests and hobbies of the target object through the historical behaviors of the target object to obtain multiple object interest features. Further, it is also necessary to mine the connection between the interests and hobbies of the target object and the content to be recommended.

[0151] S22: The server performs feature fusion based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the obtained object interest features to obtain the target fusion feature;

[0152] The content to be recommended can be subscription messages and video messages in an instant messaging application, or products and live broadcast rooms in a shopping application, or videos in a video application, which will not be listed one by one here. In the embodiments of the present application, the subscription message is mainly used as an example for illustration. As Figure 6 shown, it is a schematic diagram of a subscription message page in the embodiments of the present application, where "Xiaohu Talks about the Economy" is a subscription number, and "Is There a Wave of Bankruptcies in the New Energy Vehicle Manufacturing Industry?" and "The Economic Impact of Population Aging" are both subscription messages.

[0153] The object attribute information of the target object may include: user ID (denoted as uin), user age, user gender, user region, the number of exposed messages of the user in the past 1 day, the number of exposed messages of the user in the past 7 days, etc.

[0154] The content attribute information of the content to be recommended is divided into three categories. First, the information related to the subscription number to which the content to be recommended belongs includes: subscription number ID (denoted as bizuin), the number of fans of the subscription number, the number of messages created by the subscription number in the past 7 days, the number of clicks to read by the subscription number in the past 7 days, etc.; second, the information related to the content itself to be recommended includes: message ID, the number of hours since the message was sent, the number of exposures of the message in the past 1 hour, the number of clicks of the message in the past 1 hour, etc.; finally, the statistical information of the cross-relationship includes: the number of exposures of the target object to the subscription number in the past 28 days, the number of clicks of the target object to the subscription number in the past 28 days, etc.

[0155] The historical recommended content ID can be obtained by combining the subscription number ID and the message ID. For example, if the subscription number ID is "005" and the message ID is "20231910", then the historical recommended content ID is "00520231910". In the historical behavior sequence, each historical recommended content ID is used to represent the corresponding historical recommended content. The historical behavior sequence can be [00520231910, 00620232124, 00720232125, 00820232768].

[0156] The object attribute information, subscription number related information, content itself related information, and statistical information of cross relationships are respectively mapped through the Embedding layer to obtain object attribute features, subscription number features, content features, and cross relationship features. The subscription number features, content features, and cross relationship features can all be referred to as the content attribute features of the content to be recommended. The content attribute features include at least one of the subscription number features, content features, and cross relationship features.

[0157] The ways of feature fusion for object attribute features, content attribute features, and each object interest feature can be operations such as concatenation (concat) and addition (add), which are not specifically limited herein.

[0158] Optionally, step S22 can be implemented as:

[0159] Based on each third preset weight, the object attribute features, content attribute features, and each object interest feature are weighted and fused to obtain the target fusion feature.

[0160] Specifically, corresponding third preset weights are respectively set for each feature according to actual needs, and the object attribute features, content attribute features, and each object interest feature are weighted and fused. As Figure 7 shown, it is a schematic diagram of a feature fusion method in an embodiment of the present application. The weight of the object attribute feature is 0.2, the weight of the content attribute feature is 0.3, the weight of object interest feature 1 is 0.2, the weight of object interest feature 2 is 0.2, and the weight of object interest feature 3 is 0.2. The target fusion feature is obtained through weighted fusion according to the weight corresponding to each feature.

[0161] The object attribute features, content attribute features, and each object interest feature can also be input into a neural network model, and the object attribute features, content attribute features, and each object interest feature are weighted and fused through the weights in the model to obtain the target fusion feature.

[0162] Optionally, in step S22, before using the model for weighted fusion, the object attribute features, content attribute features, historical behavior features, and browsing sequence features generated based on each browsing duration are concatenated to obtain the concatenated feature, and then based on each third preset weight, the concatenated feature and each object interest feature are weighted and fused to obtain the target fusion feature.

[0163] S23: The server makes a prediction based on the target fusion feature to obtain the predicted value of the content to be recommended;

[0164] Among them, the predicted value of the content to be recommended can be the click-through rate, conversion rate, conversion rate after click, etc. In the embodiment of the present application, the predicted value is mainly taken as the click-through rate for illustration.

[0165] For example, the target fusion feature is input into a simple Multilayer Perceptron (MLP) model for click-through rate prediction. If the MLP has a linear-relu-linear structure, an estimated value of the click-through rate is obtained.

[0166] S24: When the estimated value meets the preset recommendation condition, the server pushes the content to be recommended to the target object.

[0167] Among them, the preset recommendation condition can be that the estimated value is greater than a preset threshold. For example, when the estimated value is greater than 0.5, the content to be recommended is pushed to the target object. The preset recommendation condition can also be to sort the estimated values, and push the content to be recommended corresponding to the top n estimated values in the sorting result to the target object.

[0168] For example, the estimated value of the content to be recommended 1 is 0.2, the estimated value of the content to be recommended 2 is 0.1, the estimated value of the content to be recommended 3 is 0.7, the estimated value of the content to be recommended 4 is 0.4, and the estimated value of the content to be recommended 5 is 0.5. The content to be recommended corresponding to the top 2 estimated values in the sorting result is the content to be recommended 3 and the content to be recommended 5, and the content to be recommended 3 and the content to be recommended 5 are pushed to the target object.

[0169] As Figure 8 shown, it is a schematic diagram of a recommendation condition in an embodiment of the present application. The target fusion feature 1 of the subscription number message 1 is input into the MLP model, and the click-through rate is obtained as 0.2. The target fusion feature 2 of the subscription number message 2 is input into the MLP model, and the click-through rate is obtained as 0.7. The target fusion feature 3 of the subscription number message 3 is input into the MLP model, and the click-through rate is obtained as 0.3. The set click-through rate threshold is 0.5, then the subscription number message 2 is pushed to the target object.

[0170] In the embodiment of the present application, during the content browsing process of the target object, based on the historical behavior features generated from the historical recommended content clicked, multiple browsing duration features are generated based on the browsing duration of the corresponding historical recommended content. Then, based on the correlation between each browsing duration feature and the historical behavior features, the corresponding feature elements in the historical behavior features are weighted respectively to obtain the corresponding object interest features. In this way, the browsing duration of the historical recommended content is introduced as supplementary information for capturing the object's interest, and the preference degree of the target object for different historical recommended content is mined through the browsing duration. The generated object interest features can fully represent the different levels of interest preferences of the target object for the content.

[0171] Further, based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the obtained object interest features of each object, feature fusion is performed to obtain the target fusion features. The obtained target fusion features can more accurately express the relationship between the interest preferences of the target object and the content to be recommended. Therefore, when making a prediction based on the target fusion features, the obtained estimated value is more accurate. When the estimated value meets the preset recommendation conditions, the content to be recommended is pushed to the target object, thereby improving the accuracy of content recommendation.

[0172] Optionally, in step S21, for a browsing duration feature, the corresponding object interest feature is obtained through the following steps S31 - S32:

[0173] S31: Based on the correlation between a browsing duration feature and historical behavior features, a weight feature is obtained;

[0174] S32: Based on the weight feature, the feature elements corresponding to the historical recommended content corresponding to a browsing duration feature in the historical behavior features are weighted to obtain the corresponding object interest feature.

[0175] Specifically, by calculating the similarity between the browsing duration feature and the historical behavior features, the correlation is obtained, and then the weight feature is obtained. There are many ways to calculate the similarity. The browsing duration feature and the historical behavior features can be feature vectors. The similarity can be obtained by taking the dot product of the browsing duration feature and the historical behavior features, or the cosine similarity, Euclidean distance (Eucledian Distance), or Manhattan distance (Manhattan Distance) can be calculated between the browsing duration feature and the historical behavior features. No specific limitation is made here.

[0176] Further, the historical behavior features are weighted by the weight feature to obtain the object interest feature. For example, the weight feature is [0.01, 0.02, …, 0.03] ∈ R 256×1 , the historical behavior features ∈ R 256×D , and the matrix multiplication of the historical behavior features is performed using the weight feature to realize the weighting of the corresponding feature elements in the historical behavior features and obtain the corresponding object interest feature.

[0177] Optionally, step S31 can be implemented as:

[0178] The feature cross of a browsing duration feature and historical behavior features is performed to obtain the behavior cross feature; the downsampling of the behavior cross feature is performed to obtain the sampling feature; the normalization process of the sampling feature is performed to obtain the weight feature.

[0179] Specifically, taking a browsing duration feature K and historical behavior features Q as an example, the dot product of Q and K is performed to obtain the behavior cross feature A″:

[0180] A″ = Q⊙K ∈ R 256×D

[0181] Perform average pooling on the behavior cross - feature A″ to obtain the sampled feature A′:

[0182] A′ = MeanPool(A″) ∈ R 256×1

[0183] Perform SoftMax operation on the sampled feature A′ (to obtain a weight vector with the sum of elements equal to 1) to obtain the attention weight (weight feature) A:

[0184] A = SoftMax(A′) ∈ R 256×1

[0185] As Figure 9 shown, it is a schematic diagram of a feature weighting method in an embodiment of the present application. Taking the browsing duration feature 1 and the historical behavior feature 1 as examples, multiply the elements at the corresponding positions in the browsing duration feature 1 and the historical behavior feature 1 to obtain the behavior cross - feature, perform average pooling on the behavior cross - feature to obtain the sampled feature, perform SoftMax operation on the sampled feature to obtain the weight feature, and use the weight feature to perform matrix multiplication on the historical behavior feature 1 to obtain the object interest vector.

[0186] Optionally, in step S22, feature fusion is performed through the following steps S41 - S42:

[0187] S41: Based on the object attribute feature, the content attribute feature, and each object interest feature, perform multiple rounds of feature cross - ing to obtain multiple information cross - features;

[0188] S42: Based on multiple information cross - features, perform feature fusion to obtain the target fusion feature.

[0189] Specifically, two features are selected for each round of feature cross - ing, and the two features selected in different rounds are not exactly the same. Taking the object attribute feature 1, the content attribute feature 1, the object interest feature 1, and the object interest feature 2 as examples, in the first round, the object attribute feature 1 and the object interest feature 1 are selected, in the second round, the object attribute feature 1 and the object interest feature 2 are selected, in the third round, the object attribute feature 1 and the content attribute feature 1 are selected, in the fourth round, the content attribute feature 1 and the object interest feature 1 are selected. Two features can continue to be selected for feature cross - ing until every two features have been feature - crossed, or the number of rounds of feature cross - ing can be set according to requirements.

[0190] Optionally, a neural network model can also be used to implement the feature fusion of object attribute features, content attribute features, and each object interest feature in step S22. For example, Logistic Regression (LR) series models and Factorization Machines (FM) series models. The object attribute features, content attribute features, and each object interest feature are input into the model, and the model will perform the above-mentioned steps of feature crossing and feature fusion to output the target fusion features.

[0191] Optionally, in one round of feature crossing in step S41, the following steps are performed:

[0192] Select two different features from the object attribute features, content attribute features, and each object interest feature; based on the inner product result of the two selected features, obtain the corresponding information crossing feature.

[0193] Specifically, the information crossing feature is obtained by taking the inner product of two features. As Figure 10 shown, it is a schematic diagram of a feature crossing process in an embodiment of the present application. For object attribute feature 1, content attribute feature 1, object interest feature 1, and object interest feature 2, the inner product of object attribute feature 1 and content attribute feature 1 obtains information crossing feature 1, the inner product of object attribute feature 1 and object interest feature 1 obtains information crossing feature 2, the inner product of object attribute feature 1 and object interest feature 2 obtains information crossing feature 3, the inner product of content attribute feature 1 and object interest feature 1 obtains information crossing feature 4, the inner product of content attribute feature 1 and object interest feature 2 obtains information crossing feature 5, and the inner product of object interest feature 1 and object interest feature 2 obtains information crossing feature 6.

[0194] Optionally, step S22 can be implemented as:

[0195] Based on each first preset weight, perform weighted fusion on multiple information crossing features to obtain a first fusion feature; based on each second preset weight, perform weighted fusion on the object attribute features, content attribute features, and each object interest feature to obtain a second fusion feature; perform feature fusion on the first fusion feature, the second fusion feature, and the preset feature to obtain the target fusion feature.

[0196] Specifically, for each information crossing feature, based on the corresponding two features, first preset weights can be set respectively. For the object attribute features, content attribute features, and each object interest feature, second preset weights can be set respectively. The preset feature is also called a global bias term and is set according to requirements.

[0197] For example, the weight of information cross - feature 1 is 0.3, the weight of information cross - feature 2 is 0.2, and the weights of object attribute feature 1, content attribute feature 1, object interest feature 1, and object interest feature 2 are all 0.1. The weight of the preset weight feature is 0.1. Based on the weights, weighted fusion is performed to obtain the target fusion feature.

[0198] As Figure 11 shown, it is a schematic structural diagram of a ranking model in an embodiment of the present application, including the following parts:

[0199] Input layer: The input data constitutes the input layer, including object attribute information, content attribute information, historical behavior sequence, and browsing duration sequence;

[0200] Embedding layer: For the discrete ID features in the input data, perform embedding mapping on them to learn a low - dimensional feature in the real - number domain (dense embeddings), and then splice it with the numerical continuous features in the input data to obtain the Embedding representation vector, which is passed into the high - order feature extraction layer.

[0201] High - order feature extraction layer: It includes a feature cross - module and a multi - key index self - attention module. Among them, the multi - key index self - attention module is used to process the historical behavior sequence and browsing duration sequence, extract user multi - interest representations U1, U2, and U3, and input them into the feature cross - module such as the FFM architecture to perform feature cross - operation with the Embedding representation vector to obtain a high - order feature vector.

[0202] Output layer: Input the high - order feature vector into a simple MLP model, and finally output the predicted click - through rate.

[0203] Conventionally, the model uses user exposure and click data as labels to supervise the predicted click - through rate output by the model, and calculates the cross - entropy loss function to train the entire model.

[0204] When applying the ranking model in the embodiment of the present application to the recommended ranking of subscription - number messages, compared with the baseline model, in group A, the baseline model is used for recommended ranking, and in group B, the ranking model in the present application is used for recommended ranking. Compared with group A, the per - capita reading duration of the official account increases by 1.270%, the per - capita number of clicked messages of subscriptions increases by 1.720%, the number of clickers increases by 0.805%, the per - capita number of article clicks increases by 1.390%, and the first - layer click increases by 1.703%. It can be seen that the per - capita number of message readings and reading duration are significantly improved, and the ranking model has better ranking ability.

[0205] As Figure 12As shown in the figure, it is a logical schematic diagram of a content recommendation method in an embodiment of the present application. Each historical recommended content clicked by the target object forms a historical behavior sequence, and the viewing duration of each historical recommended content forms a viewing duration sequence. After mapping the object attribute information, content attribute information, historical behavior sequence, and viewing duration sequence, object attribute features, content attribute features, historical behavior features, and viewing sequence features are obtained. By dividing the viewing duration sequence and based on the obtained duration sequence and historical behavior features, object interest feature 1, object interest feature 2, and object interest feature 3 are obtained. After splicing the object attribute features, content attribute features, historical behavior features, and viewing sequence features, the obtained splicing features are feature-fused with object interest feature 1, object interest feature 2, and object interest feature 3 to obtain the target fusion feature. Based on the target fusion feature, a click-through rate prediction is performed, and the click-through rate prediction value is obtained as 0.6, which is greater than the click-through rate threshold of 0.5. Then, the content to be recommended is pushed to the target object.

[0206] Based on the same inventive concept, an embodiment of the present application further provides a content recommendation device. As Figure 13 shown, it is a structural schematic diagram of the content recommendation device 1300, which may include:

[0207] A processing unit 1301, configured to weight the feature elements corresponding to the historical recommended content corresponding to each viewing duration feature in the historical behavior features respectively based on the correlation between each viewing duration feature of the target object and the historical behavior features, so as to obtain corresponding object interest features; the historical behavior features are generated based on each historical recommended content clicked by the target object, and each viewing duration feature is generated based on the viewing duration of the corresponding historical recommended content;

[0208] A fusion unit 1302, configured to perform feature fusion based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the obtained object interest features to obtain a target fusion feature;

[0209] A prediction unit 1303, configured to perform a prediction based on the target fusion feature to obtain a prediction value of the content to be recommended;

[0210] A push unit 1304, configured to push the content to be recommended to the target object when the prediction value meets a preset recommendation condition.

[0211] In an embodiment of the present application, during the content browsing process of the target object, based on the historical recommended contents clicked to generate historical behavior features, multiple browsing duration features are generated based on the browsing durations of the corresponding historical recommended contents. Then, based on the correlations between the respective browsing duration features and the historical behavior features, the feature elements corresponding to the historical behavior features are weighted respectively to obtain corresponding object interest features. In this way, the browsing duration of the historical recommended content is introduced as supplementary information for capturing the object's interest, and the preference degrees of the target object for different historical recommended contents are mined through the browsing duration. The generated object interest features can fully represent the different levels of interest preferences of the target object for the content.

[0212] Further, based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the obtained object interest features, feature fusion is performed to obtain a target fusion feature. The obtained target fusion feature can more accurately express the relationship between the interest preference of the target object and the content to be recommended. Therefore, when making a prediction based on the target fusion feature, the obtained predicted value is more accurate. When the predicted value meets the preset recommendation condition, the content to be recommended is pushed to the target object, thereby improving the accuracy of content recommendation.

[0213] Optionally, the processing unit 1301 is specifically configured to:

[0214] For each browsing duration feature, the following operations are respectively performed:

[0215] Based on the correlation between a browsing duration feature and the historical behavior features, a weight feature is obtained;

[0216] Based on the weight feature, the feature elements corresponding to the historical recommended content corresponding to a browsing duration feature in the historical behavior features are weighted to obtain corresponding object interest features.

[0217] Optionally, the processing unit 1301 is specifically configured to:

[0218] Perform feature crossing on a browsing duration feature and the historical behavior features to obtain behavior crossing features;

[0219] Perform downsampling on the behavior crossing features to obtain sampling features;

[0220] Perform normalization processing on the sampling features to obtain weight features.

[0221] Optionally, the apparatus further includes a partitioning unit 1305, configured to obtain each browsing duration feature in the following manner:

[0222] Based on at least one endpoint value, partition the browsing durations corresponding to the respective historical recommended contents to obtain corresponding duration sets;

[0223] For each of the obtained multiple duration sets, perform the following operations respectively:

[0224] For a duration set, use the viewing durations corresponding to each historical recommendation content, excluding the viewing durations included in one duration set, as candidate viewing durations, and update each obtained candidate viewing duration to a preset value;

[0225] Based on the viewing durations included in one duration set and the updated candidate viewing durations, obtain corresponding viewing duration features.

[0226] Optionally, at least one endpoint value includes a first endpoint value and a second endpoint value, and the multiple duration sets include a first duration set, a second duration set, and a third duration set;

[0227] Then the partitioning unit 1305 is specifically configured to:

[0228] For each viewing duration, partition the viewing durations less than the first endpoint value into the first duration set, partition the viewing durations greater than the second endpoint value into the second duration set, and partition the viewing durations not less than the first endpoint value and not greater than the second endpoint value into the third duration set.

[0229] Optionally, among the first ratio and the second ratio obtained based on each viewing duration, the first ratio is less than the second ratio; wherein, the first ratio is the ratio of the number of viewing durations less than the first endpoint value to the number of all viewing durations, and the second ratio is the ratio of the number of viewing durations less than the second endpoint value to the number of all viewing durations.

[0230] Optionally, the fusion unit 1302 is specifically configured to:

[0231] Based on the object attribute features, content attribute features, and each object interest feature, perform multiple rounds of feature cross - intersection to obtain multiple information cross - intersection features, where the two features selected in different rounds are not completely the same;

[0232] Based on the multiple information cross - intersection features, perform feature fusion to obtain the target fusion feature.

[0233] Optionally, the fusion unit 1302 is specifically configured to:

[0234] In one round of feature cross - intersection, select two different features from the object attribute features, content attribute features, and each object interest feature;

[0235] Based on the inner - product result of the two selected features, obtain the corresponding information cross - intersection feature.

[0236] Optionally, the fusion unit 1302 is specifically configured to:

[0237] Based on each first preset weight, perform weighted fusion on multiple information cross - features to obtain a first fusion feature;

[0238] Based on each second preset weight, perform weighted fusion on object attribute features, content attribute features, and each object interest feature to obtain a second fusion feature;

[0239] Perform feature fusion on the first fusion feature, the second fusion feature, and the preset feature to obtain a target fusion feature.

[0240] Optionally, the fusion unit 1302 is specifically configured to:

[0241] Based on each third preset weight, perform weighted fusion on object attribute features, content attribute features, and each object interest feature to obtain a target fusion feature.

[0242] For the convenience of description, the above - mentioned parts are divided into each module (or unit) according to functions and described separately. Of course, when implementing this application, the functions of each module (or unit) can be implemented in the same or multiple software or hardware.

[0243] In the embodiments of the present 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 the overall module or unit that includes the function of this module or unit.

[0244] Those skilled in the art of the relevant technical field can understand that various aspects of the present application can be implemented as a system, a method, or a program product. Therefore, various aspects of the present application can be specifically implemented in the following forms, namely: a complete hardware implementation manner, a complete software implementation manner (including firmware, microcode, etc.), or an implementation manner combining hardware and software aspects, which can be collectively referred to as "circuit", "module", or "system" here.

[0245] Based on the same inventive concept as the above - mentioned method embodiment, an electronic device is also provided in the embodiments of the present application. In one embodiment, the electronic device can be a server, such as Figure 1 the server shown. In this embodiment, the structure of the electronic device can be as Figure 14 shown, including a memory 1401, a communication module 1403, and one or more processors 1402.

[0246] A memory 1401 for storing computer programs executed by a processor 1402. The memory 1401 mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system and programs required to run the instant messaging function, etc.; the data storage area can store various instant messaging information and operation instruction sets, etc.

[0247] The memory 1401 can be a volatile memory, such as a random-access memory (RAM); the memory 1401 can also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or the memory 1401 is any other medium that can be used to carry or store a desired computer program in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 1401 can be a combination of the above memories.

[0248] The processor 1402 can include one or more central processing units (CPUs) or be a digital processing unit, etc. The processor 1402 is used to implement the above content recommendation method when calling the computer program stored in the memory 1401.

[0249] A communication module 1403 is used to communicate with terminal devices and other servers.

[0250] In the embodiments of the present application, the specific connection medium between the above-mentioned memory 1401, communication module 1403 and processor 1402 is not limited. In the embodiments of the present application Figure 14 it is described that the memory 1401 and the processor 1402 are connected through a bus 1404. The bus 1404 is described in thick lines in Figure 14 The connection manners between other components are only for illustrative purposes and are not limited thereto. The bus 1404 can be divided into an address bus, a data bus, a control bus, etc. For the convenience of description, Figure 14 it is only described by a thick line in

[0251] The memory 1401 stores a computer storage medium, and the computer storage medium stores computer-executable instructions. The computer-executable instructions are used to implement the content recommendation method of the embodiments of the present application. The processor 1402 is used to execute the above content recommendation method, as Figure 2 shown.

[0252] In another embodiment, the electronic device may also be other electronic devices, such as Figure 1 the terminal device shown. In this embodiment, the structure of the electronic device may be as Figure 15 shown, including components such as a communication component 1510, a memory 1520, a display unit 1530, a camera 1540, a sensor 1550, an audio circuit 1560, a Bluetooth module 1570, a processor 1580, etc.

[0253] The communication component 1510 is used to communicate with the server. In some embodiments, it may include a Wireless Fidelity (WiFi) module. The WiFi module belongs to short - range wireless transmission technology, and the electronic device can help users send and receive information through the WiFi module.

[0254] The memory 1520 can be used to store software programs and data. The processor 1580 executes various functions and data processing of the terminal device by running the software programs or data stored in the memory 1520. The memory 1520 may include high - speed random access memory, and may also include non - volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid - state storage devices. The memory 1520 stores an operating system that enables the terminal device to run. In this application, the memory 1520 can store the operating system and various application programs, and can also store the computer program for implementing the content recommendation method of the embodiments of this application.

[0255] The display unit 1530 can also be used to display the information input by the user or the information provided to the user, as well as the graphical user interface (GUI) of various menus of the terminal device. Specifically, the display unit 1530 may include a display screen 1532 disposed on the front of the terminal device. Among them, the display screen 1532 can be configured in the form of a liquid crystal display, a light - emitting diode, etc. The display unit 1530 can be used to display the content recommendation user interface in the embodiments of this application, etc.

[0256] The display unit 1530 can also be used to receive input digital or character information, and generate signal inputs related to the user settings and function control of the terminal device. Specifically, the display unit 1530 may include a touch screen 1531 disposed on the front of the terminal device, which can collect touch operations of the user on or near it, such as clicking buttons, dragging scroll boxes, etc.

[0257] Among them, the touch screen 1531 can cover the display screen 1532, or the touch screen 1531 and the display screen 1532 can be integrated to implement the input and output functions of the terminal device. After integration, it can be simply called a touch display screen. In this application, the display unit 1530 can display application programs and corresponding operation steps.

[0258] The camera 1540 can be used to capture static images, and users can post comments on the images captured by the camera 1540 through an application. There can be one or more cameras 1540. An object generates an optical image through a lens and projects it onto a photosensitive element. The photosensitive element can be a charge coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS) phototransistor. The photosensitive element converts the optical signal into an electrical signal, and then transmits the electrical signal to the processor 1580 to be converted into a digital image signal.

[0259] The terminal device may further include at least one sensor 1550, such as an acceleration sensor 1551, a distance sensor 1552, a fingerprint sensor 1553, and a temperature sensor 1554. The terminal device may also be configured with other sensors such as a gyroscope, a barometer, a hygrometer, a thermometer, an infrared sensor, a light sensor, and a motion sensor.

[0260] The audio circuit 1560, the speaker 1561, and the microphone 1562 can provide an audio interface between the user and the terminal device. The audio circuit 1560 can transmit the electrical signal converted from the received audio data to the speaker 1561, and the speaker 1561 converts it into a sound signal for output. The terminal device may also be configured with a volume button for adjusting the volume of the sound signal. On the other hand, the microphone 1562 converts the collected sound signal into an electrical signal, which is received by the audio circuit 1560 and converted into audio data, and then the audio data is output to the communication component 1510 to be sent to, for example, another terminal device, or the audio data is output to the memory 1520 for further processing.

[0261] The Bluetooth module 1570 is used to interact with other Bluetooth devices having a Bluetooth module through the Bluetooth protocol. For example, the terminal device can establish a Bluetooth connection with a wearable electronic device (such as a smart watch) that also has a Bluetooth module through the Bluetooth module 1570 to perform data interaction.

[0262] The processor 1580 is the control center of the terminal device, connecting various parts of the entire terminal through various interfaces and circuits. By running or executing software programs stored in the memory 1520 and invoking data stored in the memory 1520, it performs various functions of the terminal device and processes data. In some embodiments, the processor 1580 may include one or more processing units; the processor 1580 may also integrate an application processor and a baseband processor. Among them, the application processor mainly processes the operating system, user interface, application programs, etc., and the baseband processor mainly processes wireless communication. It can be understood that the above baseband processor may not be integrated into the processor 1580. In this application, the processor 1580 can run the operating system, application programs, user interface display and touch response, as well as the content recommendation method of the embodiments of this application. In addition, the processor 1580 is coupled to the display unit 1530.

[0263] In some possible implementation manners, each aspect of the content recommendation method provided in this application can also be implemented in the form of a program product, which includes a computer program. When the program product runs on an electronic device, the computer program is used to cause the electronic device to execute the steps in the content recommendation method according to various exemplary embodiments of this application described above in this specification. For example, the electronic device can execute the steps as shown in Figure 2 shown.

[0264] The program product can adopt any combination of one or more readable media. The readable media can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0265] The program product of the embodiments of this application can adopt a portable compact disc read-only memory (CD-ROM) and include a computer program, and can run on an electronic device. However, the program product of this application is not limited to this. In this document, the readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with a command execution system, apparatus, or device.

[0266] A readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, in which a readable computer program is carried. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium can also be any readable medium other than a readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with a command execution system, apparatus, or device.

[0267] The computer program contained on the readable medium can be transmitted with any appropriate medium, including but not limited to wireless, wired, optical fiber cable, RF, etc., or any suitable combination of the above.

[0268] The computer program for performing the operations of the present application can be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and also including conventional procedural programming languages such as the "C" language or similar programming languages. The computer program can be executed entirely on the user's electronic device, partially on the user's device, executed as an independent software package, partially on the user's electronic device and partially on a remote electronic device, or entirely on a remote electronic device or server. In the case of a remote electronic device, the remote electronic device can be connected to the user's electronic device through any type of network including a local area network (LAN) or a wide area network (WAN), or can be connected to an external electronic device (e.g., connected through the Internet using an Internet service provider).

[0269] It should be noted that although several units or subunits of the apparatus are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more of the above-mentioned units can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0270] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, some steps can be omitted, multiple steps can be combined into one step for execution, and / or one step can be decomposed into multiple steps for execution.

[0271] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable computer programs.

[0272] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0273] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0274] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0275] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present application.

[0276] Obviously, those skilled in the art can make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalent technologies, this application is also intended to include these modifications and variations.

Claims

1. A content recommendation method, characterized in that, The method includes: Based on the correlation between each browsing duration feature of the target object and the historical behavior features, respectively weight the feature elements corresponding to the historical recommended content corresponding to each browsing duration feature in the historical behavior features to obtain corresponding object interest features; the historical behavior features are generated based on each historical recommended content clicked by the target object, and each browsing duration feature is generated based on the browsing duration of the corresponding historical recommended content; Based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the obtained object interest features, perform feature fusion to obtain target fusion features; Based on the target fusion features, make a prediction to obtain an estimated value of the content to be recommended; When the estimated value meets the preset recommendation condition, push the content to be recommended to the target object.

2. The method according to claim 1, wherein The step of, based on the correlation between each browsing duration feature of the target object and the historical behavior features, respectively weight the feature elements corresponding to the historical recommended content corresponding to each browsing duration feature in the historical behavior features to obtain corresponding object interest features, includes: For each of the browsing duration features, respectively perform the following operations: Based on the correlation between one browsing duration feature and the historical behavior features, obtain a weight feature; Based on the weight feature, weight the feature elements corresponding to the historical recommended content corresponding to the one browsing duration feature in the historical behavior features to obtain corresponding object interest features.

3. The method according to claim 2, wherein The step of, based on the correlation between one browsing duration feature and the historical behavior features, obtain a weight feature, includes: Perform feature crossing on the one browsing duration feature and the historical behavior features to obtain a behavior crossing feature; Perform downsampling on the behavior crossing feature to obtain a sampled feature; Perform normalization processing on the sampled feature to obtain the weight feature.

4. The method according to claim 1, wherein The various browsing duration features are obtained by the following method: Based on at least one endpoint value, divide the browsing duration corresponding to each historical recommended content to obtain corresponding duration sets; For each of the obtained multiple duration sets, respectively perform the following operations: For one duration set, use the browsing durations corresponding to each historical recommended content except the browsing durations included in the one duration set as candidate browsing durations, and update the obtained candidate browsing durations to preset values; Based on the browsing durations included in the one duration set and the updated candidate browsing durations, obtain corresponding browsing duration features.

5. The method according to claim 4, characterized in that, The at least one endpoint value includes a first endpoint value and a second endpoint value, and the multiple duration sets include a first duration set, a second duration set, and a third duration set; Then the step of, based on at least one endpoint value, divide the browsing duration of each historical recommended content to obtain multiple duration sets, includes: For each of the browsing durations, the browsing durations less than the first endpoint value are classified into the first duration set, the browsing durations greater than the second endpoint value are classified into the second duration set, and the browsing durations not less than the first endpoint value and not greater than the second endpoint value are classified into the third duration set.

6. The method according to claim 5, wherein Among the first ratio and the second ratio obtained based on the respective browsing durations, the first ratio is less than the second ratio; wherein, the first ratio is the ratio of the number of browsing durations less than the first endpoint value to the number of all the browsing durations among the respective browsing durations, and the second ratio is the ratio of the number of browsing durations less than the second endpoint value to the number of all the browsing durations among the respective browsing durations.

7. The method according to any one of claims 1 to 6, characterized in that The feature fusion based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the respective object interest features to obtain the target fusion features includes: Performing multiple rounds of feature crossing based on the object attribute features, the content attribute features, and the respective object interest features to obtain multiple information crossing features, where the two features selected in different rounds are not completely the same; Performing feature fusion based on the multiple information crossing features to obtain the target fusion features.

8. The method according to claim 7, wherein The performing multiple rounds of feature crossing based on the object attribute features, the content attribute features, and the respective object interest features to obtain multiple information crossing features includes: In one round of feature crossing, two different features are selected from the object attribute features, the content attribute features, and the respective object interest features; Based on the inner product result of the two selected features, the corresponding information crossing feature is obtained.

9. The method according to claim 7, wherein The performing feature fusion based on the multiple information crossing features to obtain the target fusion features includes: Performing weighted fusion on the multiple information crossing features based on respective first preset weights to obtain a first fusion feature; Performing weighted fusion on the object attribute features, the content attribute features, and the respective object interest features based on respective second preset weights to obtain a second fusion feature; Performing feature fusion on the first fusion feature, the second fusion feature, and a preset feature to obtain the target fusion features.

10. The method according to any one of claims 1 to 6, characterized in that, The feature fusion based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the respective object interest features to obtain the target fusion features includes: Performing weighted fusion on the object attribute features, the content attribute features, and the respective object interest features based on respective third preset weights to obtain the target fusion features.

11. A content recommendation device, characterized in that, Including: A processing unit, configured to, based on the correlation between each browsing duration feature of the target object and the historical behavior feature, respectively weight the feature elements corresponding to the historical recommended content corresponding to each browsing duration feature in the historical behavior feature to obtain corresponding object interest features; the historical behavior feature is generated based on each historical recommended content clicked by the target object, and each browsing duration feature is generated based on the browsing duration of the corresponding historical recommended content. A fusion unit, configured to perform feature fusion based on the object attribute features of the target object, the content attribute features of the content to be recommended, and the obtained object interest features, so as to obtain target fusion features; A prediction unit, configured to perform a prediction based on the target fusion features to obtain an estimated value of the content to be recommended; A push unit, configured to push the content to be recommended to the target object when the estimated value meets a preset recommendation condition.

12. An electronic device, characterized in that, It includes a processor and a memory. Among them, the memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, It includes a computer program. When the computer program runs on an electronic device, the computer program is used to cause the electronic device to execute the steps of any one of claims 1 to 10.

14. A computer program product, characterized in that, It includes a computer program, and the computer program is stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, so that the electronic device executes the steps of any one of claims 1 to 10.

Images