News recommendation method, device, media server and readable storage medium

Abstract

The present application provides a news recommendation method, device, media server and readable storage medium, and relates to the technical field of information recommendation. The present application extracts news keywords from the news content currently read by the target user to update the interest keywords stored in the interest database of the target user, and determines the user interest feature vector based on all the keywords included in the updated interest database, then extracts the news content feature vector from the news to be recommended, and calculates the click probability of the news to be recommended being clicked by the target user based on the user interest feature vector and the news content feature vector, and then directly recommends the news to be recommended to the target user when the click probability is greater than or equal to the preset probability threshold, thereby reducing the amount of computing resource consumption in the news recommendation process, improving the news recommendation efficiency, and ensuring that the recommended news content matches the user interest change trend in real time, thereby realizing the personalized news recommendation function.

Description

Technical Field

[0001] The present application relates to the technical field of information recommendation, and in particular to a news recommendation method, device, media server and readable storage medium. Background Art

[0002] With the continuous development of Internet technology, various types of information are growing exponentially, and it is becoming increasingly difficult for people to find content of interest from massive amounts of information. Therefore, appropriate information recommendation algorithms are gradually developed for various industries to recommend appropriate information to users, so as to improve the user's information reading experience. However, in the process of news recommendation, existing news recommendation schemes still have major defects in recommendation efficiency, and / or cannot ensure that the recommended news matches the changing trend of user interests in real time to achieve personalized news recommendation. Summary of the invention

[0003] In view of this, the purpose of the present application is to provide a news recommendation method, device, media server and readable storage medium, which can improve the efficiency of news recommendation while realizing the personalized news recommendation function by recommending news content that matches the changing trends of user interests in real time to users.

[0004] In order to achieve the above purpose, the technical solution adopted in the embodiment of the present application is as follows:

[0005] In a first aspect, an embodiment of the present application provides a news recommendation method, the method comprising:

[0006] Extract keywords from the news content currently read by the target user to obtain corresponding news keywords;

[0007] Adjusting the content of interest keywords stored in the interest database corresponding to the target user according to the news keywords, and adding the news keywords to the interest database;

[0008] Calculating a current user interest feature vector of the target user according to all keywords included in the interest database;

[0009] Extracting news word segmentation vectors of the news to be recommended, and calculating news content feature vectors of the news to be recommended based on the extracted news word segmentation vectors;

[0010] Calculating the click probability of the to-be-recommended news being clicked by the target user according to the user interest feature vector and the news content feature vector;

[0011] When the click probability is greater than or equal to a preset probability threshold, the news to be recommended is recommended to the target user.

[0012] In an optional implementation, the step of adjusting the content of the interest keywords stored in the interest database corresponding to the target user according to the news keywords and adding the news keywords to the interest database includes:

[0013] Calculating feature similarity between each of the news keywords and each of the interest keywords in the interest database;

[0014] The first interest keyword and the interest renewal rate of the first interest keyword stored in the interest database are deleted, and the interest renewal rates of the remaining interest keywords in the interest database are updated using a preset time migration rate, wherein the first interest keyword is an interest keyword whose feature similarity with the news keyword is greater than a preset similarity threshold, and the updated interest renewal rate is equal to the product of the preset time migration rate and the interest renewal rate before the update;

[0015] Deleting the second interest keyword and the interest freshness rate of the second interest keyword from the remaining interest keywords in the interest database, wherein the second interest keyword is an interest keyword whose corresponding interest freshness rate is less than a preset freshness rate threshold;

[0016] For each of the news keywords, an interest freshness rate with a value of 1 is configured for the news keyword, and the news keyword and its interest freshness rate are added to the interest database.

[0017] In an optional implementation, the step of calculating the current user interest feature vector of the target user according to all keywords included in the interest database includes:

[0018] Performing word vector conversion on each keyword in the interest database to obtain corresponding keyword vectors;

[0019] The interest newness rate of each keyword in the interest database is used as a weight, and the obtained keyword vectors are weighted and summed to obtain the user interest feature vector.

[0020] In an optional implementation, the step of extracting news word segmentation vectors of the news to be recommended includes:

[0021] Performing word segmentation processing on the news content of the news to be recommended to obtain the word segmentation result of the news to be recommended;

[0022] The word segmentation result is processed to remove stop words to obtain news word segmentations corresponding to the news to be recommended;

[0023] The obtained news word segmentation is converted into a word vector to obtain a corresponding news word segmentation vector.

[0024] In an optional implementation, the step of calculating the news content feature vector of the news to be recommended based on the extracted news word segmentation vector includes:

[0025] Input the news word segmentation vector into the attention mechanism model;

[0026] The attention mechanism model is called at the word level attention level and the sentence level attention level to perform feature fusion on the news word segmentation vector to obtain the news content feature vector.

[0027] In a second aspect, an embodiment of the present application provides a news recommendation device, the device comprising:

[0028] The keyword extraction module is used to extract keywords from the news content currently read by the target user to obtain the corresponding news keywords;

[0029] A database updating module, configured to adjust the content of the interest keywords stored in the interest database corresponding to the target user according to the news keywords, and add the news keywords to the interest database;

[0030] A feature vector calculation module, used to calculate the current user interest feature vector of the target user according to all keywords included in the interest database;

[0031] The feature vector calculation module is further used to extract the news word segmentation vector of the news to be recommended, and calculate the news content feature vector of the news to be recommended based on the extracted news word segmentation vector;

[0032] A click probability calculation module, used to calculate the click probability of the recommended news being clicked by the target user according to the user interest feature vector and the news content feature vector;

[0033] The news recommendation module is used to recommend the news to be recommended to the target user when the click probability is greater than or equal to a preset probability threshold.

[0034] In an optional implementation manner, the database updating module adjusts the content of the interest keywords stored in the interest database corresponding to the target user according to the news keywords, and adds the news keywords to the interest database, including:

[0035] Calculating feature similarity between each of the news keywords and each of the interest keywords in the interest database;

[0036] The first interest keyword and the interest renewal rate of the first interest keyword stored in the interest database are deleted, and the interest renewal rates of the remaining interest keywords in the interest database are updated using a preset time migration rate, wherein the first interest keyword is an interest keyword whose feature similarity with the news keyword is greater than a preset similarity threshold, and the updated interest renewal rate is equal to the product of the preset time migration rate and the interest renewal rate before the update;

[0037] Deleting the second interest keyword and the interest freshness rate of the second interest keyword from the remaining interest keywords in the interest database, wherein the second interest keyword is an interest keyword whose corresponding interest freshness rate is less than a preset freshness rate threshold;

[0038] For each of the news keywords, an interest freshness rate with a value of 1 is configured for the news keyword, and the news keyword and its interest freshness rate are added to the interest database.

[0039] In an optional implementation manner, the feature vector calculation module calculates the current user interest feature vector of the target user according to all keywords included in the interest database, including:

[0040] Performing word vector conversion on each keyword in the interest database to obtain corresponding keyword vectors;

[0041] The interest newness rate of each keyword in the interest database is used as a weight, and the obtained keyword vectors are weighted and summed to obtain the user interest feature vector.

[0042] In a third aspect, an embodiment of the present application provides a media server, comprising a processor and a memory, wherein the memory stores a computer program that can be executed by the processor, and the processor can execute the computer program to implement the news recommendation method described in any one of the aforementioned implementations.

[0043] In a fourth aspect, an embodiment of the present application provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the news recommendation method described in any one of the aforementioned implementations is implemented.

[0044] The beneficial effects of the embodiments of the present application include the following:

[0045] The present application extracts news keywords from the news content currently read by the target user, and after adjusting the content of the interest keywords stored in the interest database of the target user by using the news keywords, adds the news keywords to the interest database, and then determines the user's current user interest feature vector based on all the keywords included in the interest database, and then extracts the corresponding news content feature vector from the news to be recommended, and calculates the click probability of the news to be recommended by the target user based on the obtained user interest feature vector and the news content feature vector, and then directly recommends the news to be recommended to the target user when the click probability is greater than or equal to a preset probability threshold, thereby reducing the amount of computing resource loss in the existing news recommendation process, improving the news recommendation efficiency, and ensuring that the recommended news content matches the user's interest change trend in real time, thereby realizing the news personalized recommendation function.

[0046] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A schematic diagram of the composition of a media server provided in an embodiment of the present application;

[0049] Figure 2 A flowchart of a news recommendation method provided in an embodiment of the present application;

[0050] Figure 3 for Figure 2 A schematic flow chart of the sub-steps included in step S220;

[0051] Figure 4 for Figure 2 A schematic flow chart of the sub-steps included in step S230;

[0052] Figure 5 for Figure 2 One of the flowchart diagrams of the sub-steps included in step S240;

[0053] Figure 6 for Figure 2 2 is a flow chart of sub-steps included in step S240;

[0054] Figure 7 A schematic diagram of the composition of a news recommendation device provided in an embodiment of the present application.

[0055] Icons: 10 - media server; 11 - memory; 12 - processor; 13 - communication unit; 100 - news recommendation device; 110 - keyword extraction module; 120 - database update module; 130 - feature vector calculation module; 140 - click probability calculation module; 150 - news recommendation module. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0057] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0058] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0059] In the description of the present application, it is to be understood that the relational terms such as the term "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "includes", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or equipment. In the absence of more restrictions, the elements defined by the sentence "comprising one..." do not exclude the existence of other identical elements in the process, method, article or equipment including the elements. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood by specific circumstances.

[0060] The applicant has found through hard research that existing news recommendation solutions are usually implemented using four approaches:

[0061] The first recommendation method is based on popularity. It usually requires ranking the recommended content according to the popularity of the news content to be recommended (such as the number of likes, comments, etc.), and recommending the top ranked content to the user. It should be noted that this recommendation method requires less user interest information, cannot make personalized recommendations for users, and the recommended content is relatively simple, which does not meet the user's information search needs.

[0062] The second recommendation method is content-based recommendation method, which usually requires selecting some historical click contents of users for manual annotation, and then using computer technology to automatically feature each user's historical click news, so as to characterize the user's profile through information retrieval and information filtering as well as the user's historical click information, and recommend the information that the user is more interested in. It should be noted that this recommendation method needs to process the text information of all historical click news of users, which consumes too much computing resources and time in the process of natural language processing, making it difficult to ensure the efficiency of news recommendation, and it cannot represent the user's interest deviation in real time.

[0063] The third recommendation method is based on collaborative filtering. It usually takes advantage of the idea of ​​"birds of a feather flock together" and uses the historical click content of similar users as an important indicator for recommendation. Therefore, this recommendation method can be divided into two categories: (1) User-based collaborative filtering recommendation method. This method scores the recommended items by finding the historical clicks of users with similar or same interests as the current user, thereby achieving the purpose of recommendation; (2) Item-based collaborative filtering method. This method recommends items similar to those they liked before to users. Through users scoring different behaviors of items, item representation vectors are constructed, and recommendations are made based on the similarity between items. It should be noted that although this recommendation method improves the recommendation efficiency to a certain extent and can discover the potential interests of the target user from the preferences of similar users, this method has a cold start problem. That is, if the target user is a new user with less historical click information, or the target user's interests are relatively unpopular, this method cannot perform effective personalized news recommendation when there are very few users with similar interests and hobbies as the target user in the big data.

[0064] The fourth recommendation idea is a model-based recommendation method, which usually requires modeling the target user and its recommended content through machine learning, converting the entities in the recommended news into more universal contextual entities, and combining the news title, title entity and contextual entity to represent the user portrait, so as to efficiently and automatically predict the user's preferences, obtain the scoring results of the recommended news, and then recommend appropriate news to the user. It should be noted that this recommendation idea requires natural processing of the text and then using similarity comparison or combining all information to input into a multi-layer perceptron for prediction. It has high demand for computing resources and is relatively time-consuming, and there is a problem of low news recommendation efficiency. At the same time, this method ignores the migration of user interests because it needs to use the same method to calculate all news. It cannot guarantee that the recommended news matches the user's interest deviation and cannot achieve personalized news recommendation.

[0065] To this end, the applicant provides a news recommendation method, device, media server and readable storage medium to improve the efficiency of news recommendation, and recommend news content that matches the changing trends of user interests in real time to users, thereby realizing a personalized news recommendation function.

[0066] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0067] Please refer to Figure 1 , Figure 1 1 is a schematic diagram of the composition of the media server 10 provided in the embodiment of the present application. In the embodiment of the present application, the media server 10 can quickly recommend news content that matches the user's interest change trend in real time to the user, thereby realizing a personalized news recommendation function.

[0068] In this embodiment, the media server 10 may include a memory 11, a processor 12, a communication unit 13, and a news recommendation device 100. The memory 11, the processor 12, and the communication unit 13 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, the memory 11, the processor 12, and the communication unit 13 may be electrically connected to each other via one or more communication buses or signal lines.

[0069] In this embodiment, the memory 11 may be, but not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. The memory 11 is used to store programs, and the processor 12 may execute the programs accordingly after receiving the execution instruction.

[0070] The memory 11 can also be used to store user accounts of multiple users, and an interest database corresponding to each user account. The interest database is used to represent the changing trend of user interests over time, and the interest database stores interest keywords used to represent the user's existing interests, and the interest freshness rate of each interest keyword. The interest freshness rate is used to represent the newness of the corresponding interest over time, representing the timeliness of the corresponding interest, and its value is less than or equal to 1, and the smaller the value, the higher the possibility that the corresponding interest is forgotten by the user.

[0071] In this embodiment, the processor 12 may be an integrated circuit chip with signal processing capability. The processor 12 may be a general-purpose processor, including at least one of a central processing unit (CPU), a graphics processing unit (GPU), and a network processor (NP). The general-purpose processor may be a microprocessor or any conventional processor, etc., which may implement or execute the methods, steps, and logic diagrams disclosed in the embodiments of the present application.

[0072] In this embodiment, the communication unit 13 is used to establish a communication connection between the media server 10 and other electronic devices through a network, and to send and receive data through the network, wherein the network includes a wired communication network and a wireless communication network. For example, the media server 10 obtains relevant information about the news currently read by the target user from the terminal device logged in with the user account of the target user through the communication unit 13, wherein the relevant information about the news includes the news title, news number, etc. of the news. The media server 10 can also push a certain news to the terminal device logged in with the user account of the target user through the communication unit 13.

[0073] In this embodiment, the news recommendation device 100 includes at least one software function module that can be stored in the memory 11 in the form of software or firmware or fixed in the operating system of the media server 10. The processor 12 can be used to execute the executable module stored in the memory 11, such as the software function module and computer program included in the news recommendation device 100. The media server 10 introduces the concept of keywords through the news recommendation device 100 to determine the current user interest characteristics of the target user, and intuitively infers the possibility of the corresponding news to be recommended being clicked and browsed by the target user in combination with the news content of the news to be recommended, and then recommends the news to be recommended to the target user for browsing when the possibility is high, thereby greatly reducing the amount of computing resource consumption in the prior art for describing the user interest migration status, improving the news recommendation efficiency, and ensuring that the news content recommended to the target user matches the user interest change trend in real time, realizing the personalized news recommendation function.

[0074] Understandably, Figure 1 The block diagram shown is only a schematic diagram of a composition of the media server 10. The media server 10 may also include Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown. Figure 1 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0075] In this application, in order to ensure that the media server 10 can effectively improve the news recommendation efficiency and ensure the personalized recommendation effect of news, this application implements the above functions by providing a news recommendation method applied to the above media server 10. The news recommendation method provided by this application is described in detail below.

[0076] Please refer to Figure 2 , Figure 2 : is a flow chart of the news recommendation method provided in the embodiment of the present application. In the embodiment of the present application, Figure 2 The specific process and steps of the news recommendation method are shown below.

[0077] Step S210: extract keywords from the news content currently read by the target user to obtain corresponding news keywords.

[0078] In this embodiment, the media server 10 can obtain relevant information about the news currently read by the target user from the terminal device with the user account of the target user logged in through the communication unit 13, and then obtain the news content currently read by the target user based on the relevant information, and use the term frequency-inverse document frequency (Term Frequency-Inverse Document Frequency, TF-IDF) algorithm to extract keywords from the obtained news content to obtain news keywords included in the news currently read by the target user. At this time, the news keywords can represent the news content features that the target user is currently interested in.

[0079] Step S220, adjusting the content of the interest keywords stored in the interest database corresponding to the target user according to the news keywords, and adding the news keywords to the interest database.

[0080] In this embodiment, after obtaining the news keywords included in the news currently read by the target user, the media server 10 can update the influence of the existing interest keywords in the interest database of the target user during the user's interest changes according to the news keywords, and then add the news keywords to the updated interest database, so that the final interest database can be matched with the user's real interests in real time.

[0081] Optionally, see Figure 3 , Figure 3 yes Figure 2 Schematic diagram of the flow of sub-steps included in step S220 of FIG. In this embodiment, step S220 may include sub-steps S221 to S224.

[0082] Sub-step S221, calculating the feature similarity between each news keyword and each interest keyword in the interest database.

[0083] In this embodiment, the media server 10 can use a word vector model (e.g., Word2vec, glove, one-hot, fastText, Bert, Elmo, etc.) to encode each interest keyword in the interest database and each currently obtained news keyword, complete the word vector conversion of each interest keyword and each currently obtained news keyword, and obtain the corresponding interest keyword vector and news keyword vector. Then, calculate the cosine similarity between each interest keyword vector and each news keyword vector, and use the obtained cosine similarity as the feature similarity between the corresponding interest keyword vector and the news keyword vector.

[0084] Sub-step S222, deleting the first interest keyword and the interest renewal rate of the first interest keyword stored in the interest database, and updating the interest renewal rates of the remaining interest keywords in the interest database using a preset time migration rate.

[0085] In this embodiment, the first interest keyword is an interest keyword in the interest database whose feature similarity with the news keyword is greater than a preset similarity threshold. When a first interest keyword is found, it indicates that the corresponding user has found a news keyword that can replace the first interest keyword in the news currently being read. The media server 10 can delete the first interest keyword and the interest freshness rate of the first interest keyword in the interest database, so that the keyword content is substantially repeated when the news keyword is added to the interest database, and the timeliness of the corresponding keyword content is extended.

[0086] When the media server 10 deletes all the first interest keywords and the interest renewal rates of the first interest keywords stored in the interest database, the preset time migration rate will be used to update the interest renewal rates of the remaining interest keywords in the interest database to adjust the influence of each remaining interest keyword in the process of user interest change. Among them, the updated interest renewal rate is equal to the product of the preset time migration rate and the interest renewal rate before the update. The preset time migration rate is a decimal value less than 1, so the updated interest renewal rate will be correspondingly smaller than the interest renewal rate before the update.

[0087] Sub-step S223, deleting the second interest keyword and the interest renewal rate of the second interest keyword from the remaining interest keywords in the interest database.

[0088] In this embodiment, the second interest keyword is an interest keyword in the interest database whose corresponding interest freshness rate is less than a preset freshness rate threshold. When a second interest keyword is found, it can be indicated that the influence of the second interest keyword in the current user interest change process is too small, and its essence cannot represent the user's current real intention to select news. At this time, the media server 10 will delete all the second interest keywords in the interest database and the interest freshness rate of each second interest keyword, so as to avoid data redundancy in the interest database and improve the adaptability of the interest database to the target user.

[0089] Sub-step S224, for each news keyword, configure an interest freshness rate with a value of 1 for the news keyword, and add the news keyword and its interest freshness rate to the interest database.

[0090] In this embodiment, when the media server 10 deletes the first interest keyword and its corresponding interest freshness rate, as well as the second interest keyword and its corresponding interest freshness rate in the interest database, it will use all the currently obtained news keywords as interest keywords to be added to the interest database, and configure an interest freshness rate of 1 for each currently obtained news keyword, and then add the news keyword and its interest freshness rate to the interest database, thereby ensuring that the final interest database substantially matches the current interest change trend of the target user.

[0091] Therefore, the media server 10 can ensure that the interest database finally obtained can match the real interests of the target user in real time and reflect the current interest change trend of the target user by executing the above sub-steps S221 to S224.

[0092] Step S230, calculating the current user interest feature vector of the target user according to all the keywords included in the interest database.

[0093] In this embodiment, after the content update of the interest database is completed, the current user interest feature vector of the target user can be obtained by performing vector fusion on all keywords included in the interest database.

[0094] Optionally, see Figure 4 , Figure 4 yes Figure 2 Schematic diagram of the flow of sub-steps included in step S230. In this embodiment, step S230 may include sub-steps S231 to S232.

[0095] Sub-step S231, converting each keyword in the interest database into a word vector to obtain the corresponding keyword vector.

[0096] In this embodiment, the media server 10 can use a word vector model (for example, Word2vec, glove, one-hot, fastText, Bert and Elmo, etc.) to encode each keyword in the interest database at this time, complete the word vector conversion of all keywords in the interest database, and obtain corresponding keyword vectors.

[0097] Sub-step S232, taking the interest newness rate of each keyword in the interest database as a weight, performing weighted summation on the obtained keyword vectors to obtain a user interest feature vector.

[0098] In this embodiment, the media server 10 may use the interest renewal rate of each keyword in the interest database as a weight, perform weighted summation on the obtained keyword vectors, and use the final weighted summation result as the current user interest feature vector of the target user.

[0099] Step S240, extracting the news word segmentation vector of the news to be recommended, and calculating the news content feature vector of the news to be recommended based on the extracted news word segmentation vector.

[0100] In this embodiment, the news to be recommended is a piece of news randomly selected by the media server 10 that may be recommended to the target user. The media server 10 extracts a news content feature vector for characterizing the essential news characteristics of the news to be recommended from the news to be recommended.

[0101] Optionally, see Figure 5 , Figure 5 yes Figure 2 One of the flowcharts of the sub-steps included in step S240 of the method. In this embodiment, the step of extracting the news word segmentation vector of the news to be recommended in step S240 may include sub-steps S241 to S243.

[0102] Sub-step S241, performing word segmentation processing on the news content of the news to be recommended to obtain the word segmentation result of the news to be recommended.

[0103] Sub-step S242, removing stop words from the word segmentation results to obtain news word segmentations corresponding to the news to be recommended.

[0104] Sub-step S243, converting the obtained news word segmentation into a word vector to obtain a corresponding news word segmentation vector.

[0105] Among them, the media server 10 can use a word vector model (for example, Word2vec, glove, one-hot, fastText, Bert and Elmo, etc.) to encode the obtained news word segmentation, complete the word vector conversion of the news word segmentation, and obtain the corresponding news word segmentation vector.

[0106] Optionally, see Figure 6 , Figure 6 yes Figure 2 2 is a flow chart of sub-steps included in step S240 of the method. In this embodiment, the step of calculating the news content feature vector of the news to be recommended according to the extracted news word segmentation vector in step S240 may include sub-steps S245 and S246.

[0107] Sub-step S245, input the news word segmentation vector into the attention mechanism model.

[0108] Sub-step S246, calls the attention mechanism model at the word level attention level and the sentence level attention level to perform feature fusion on the news word segmentation vector to obtain the news content feature vector.

[0109] In this embodiment, the attention mechanism model may include a word level sub-model and a sentence level sub-model, the output end of the word level sub-model is connected to the input end of the sentence level sub-model, the input end of the word level sub-model is used to receive the news word segmentation vector, and the output end of the sentence level sub-model is used to output the news content feature vector. Among them, the word level sub-model and the sentence level sub-model can be constructed using any network structure of a bidirectional gated recurrent unit, a unidirectional gated recurrent unit, a bidirectional long short-term memory network, and a unidirectional long short-term memory network, and the coordination between the word level sub-model and the sentence level sub-model can be realized by using any attention mechanism such as a hierarchical attention mechanism, a self-attention mechanism, and a multi-head self-attention mechanism.

[0110] Please refer again Figure 2 , step S250, calculating the click probability of the recommended news being clicked by the target user based on the user interest feature vector and the news content feature vector.

[0111] In this embodiment, the media server 10 may obtain the click probability of the target user clicking on the news to be recommended by performing vector dot product of the obtained user interest feature vector and the news content feature vector.

[0112] Step S260: When the click probability is greater than or equal to the preset probability threshold, the news to be recommended is recommended to the target user.

[0113] In this embodiment, when the media server 10 calculates that the click probability of the news to be recommended under the current user interest feature vector of the target user is greater than or equal to the preset probability threshold, it can be indicated that the news to be recommended currently selected by the media server 10 matches the current interest change trend of the target user in real time. The media server 10 will correspondingly send the access connection of the news to be recommended to the terminal device used by the target user, thereby completing the entire personalized news recommendation function.

[0114] Therefore, the media server 10 can introduce the concept of keywords to determine the current user interest characteristics of the target user by executing the above steps S210 to S260, and intuitively infer the possibility of the corresponding news to be recommended being clicked and browsed by the target user in combination with the news content of the news to be recommended, and then recommend the news to be recommended to the target user for browsing when the possibility is high, thereby greatly reducing the amount of computing resource consumption in describing the user interest migration status in the prior art, improving the news recommendation efficiency, and ensuring that the news content recommended to the target user matches the user interest change trend in real time, thereby realizing the personalized news recommendation function.

[0115] In this application, in order to ensure that the news recommendation device 100 included in the media server 10 can be implemented normally, this application implements its functions by dividing the news recommendation device 100 into functional modules. The specific components of the news recommendation device 100 provided by this application are described below.

[0116] Please refer to Figure 7 , Figure 7 1 is a schematic diagram of the components of the news recommendation device 100 provided in the embodiment of the present application. In the embodiment of the present application, the news recommendation device 100 may include a keyword extraction module 110, a database update module 120, a feature vector calculation module 130, a click probability calculation module 140 and a news recommendation module 150.

[0117] The keyword extraction module 110 is used to extract keywords from the news content currently read by the target user to obtain corresponding news keywords.

[0118] The database updating module 120 is used to adjust the content of the interest keywords stored in the interest database corresponding to the target user according to the news keywords, and add the news keywords to the interest database.

[0119] The feature vector calculation module 130 is used to calculate the current user interest feature vector of the target user according to all keywords included in the interest database.

[0120] The feature vector calculation module 130 is further used to extract news word segmentation vectors of the news to be recommended, and calculate news content feature vectors of the news to be recommended based on the extracted news word segmentation vectors.

[0121] The click probability calculation module 140 is used to calculate the click probability of the recommended news being clicked by the target user based on the user interest feature vector and the news content feature vector.

[0122] The news recommendation module 150 is used to recommend the news to be recommended to the target user when the click probability is greater than or equal to a preset probability threshold.

[0123] The database updating module 120 adjusts the content of the interest keywords stored in the interest database corresponding to the target user according to the news keywords, and adds the news keywords to the interest database, including:

[0124] Calculate the feature similarity between each news keyword and each interest keyword in the interest database;

[0125] The first interest keyword and the interest freshness rate of the first interest keyword stored in the interest database are deleted, and the interest freshness rates of the remaining interest keywords in the interest database are updated using a preset time migration rate, wherein the first interest keyword is an interest keyword whose feature similarity with the news keyword is greater than a preset similarity threshold, and the updated interest freshness rate is equal to the product of the preset time migration rate and the interest freshness rate before the update;

[0126] Deleting the second interest keyword and the interest freshness rate of the second interest keyword from the remaining interest keywords in the interest database, wherein the second interest keyword is an interest keyword whose corresponding interest freshness rate is less than a preset freshness rate threshold;

[0127] For each news keyword, an interest freshness rate with a value of 1 is configured for the news keyword, and the news keyword and its interest freshness rate are added to the interest database.

[0128] The feature vector calculation module 130 calculates the current user interest feature vector of the target user according to all keywords included in the interest database, including:

[0129] Perform word vector conversion on each keyword in the interest database to obtain the corresponding keyword vector;

[0130] The interest newness rate of each keyword in the interest database is used as a weight, and the obtained keyword vectors are weighted and summed to obtain the user interest feature vector.

[0131] The feature vector calculation module 130 extracts the news word segmentation vector of the news to be recommended, including:

[0132] Perform word segmentation processing on the news content of the news to be recommended to obtain the word segmentation result of the news to be recommended;

[0133] Remove stop words from the word segmentation results to obtain news word segmentations corresponding to the news to be recommended;

[0134] The obtained news word segmentation is converted into a word vector to obtain the corresponding news word segmentation vector.

[0135] The feature vector calculation module 130 calculates the news content feature vector of the news to be recommended according to the extracted news word segmentation vector, including:

[0136] Input the news word segmentation vector into the attention mechanism model;

[0137] The attention mechanism model is called to perform feature fusion on the news word segmentation vector at the word level attention level and the sentence level attention level to obtain a news content feature vector.

[0138] It should be noted that the basic principle and technical effects of the news recommendation device 100 provided in the embodiment of the present application are the same as those of the aforementioned news recommendation method. For the sake of brief description, for parts not mentioned in this embodiment, reference may be made to the description of the aforementioned news recommendation method.

[0139] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic, for example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to the embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or the flowchart, and the combination of boxes in the block diagram and / or the flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0140] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0141] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a readable storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned readable storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0142] In summary, in a news recommendation method, device, media server and readable storage medium provided by the present application, the present application extracts news keywords from the news content currently read by the target user, and after using the news keywords to adjust the content of the interest keywords stored in the interest database of the target user, adds the news keywords to the interest database, and then determines the user's current user interest feature vector based on all the keywords included in the interest database, and then extracts the corresponding news content feature vector from the news to be recommended, and calculates the click probability of the news to be recommended by the target user based on the obtained user interest feature vector and the news content feature vector, and then directly recommends the news to be recommended to the target user when the click probability is greater than or equal to a preset probability threshold, thereby reducing the amount of computing resource loss in the existing news recommendation process, improving the news recommendation efficiency, and ensuring that the recommended news content matches the user's interest change trend in real time, thereby realizing the personalized news recommendation function.

[0143] The above are only various implementations of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A news recommendation method, characterized in that: The method comprises: Extract keywords from the news content currently read by the target user to obtain corresponding news keywords; Adjusting the content of interest keywords stored in the interest database corresponding to the target user according to the news keywords, and adding the news keywords to the interest database; Calculating a current user interest feature vector of the target user according to all keywords included in the interest database; Extracting news word segmentation vectors of the news to be recommended, and calculating news content feature vectors of the news to be recommended based on the extracted news word segmentation vectors; Calculating the click probability of the recommended news being clicked by the target user according to the user interest feature vector and the news content feature vector; When the click probability is greater than or equal to a preset probability threshold, recommending the news to be recommended to the target user; The step of adjusting the content of the interest keywords stored in the interest database corresponding to the target user according to the news keywords and adding the news keywords to the interest database includes: Calculating feature similarity between each of the news keywords and each of the interest keywords in the interest database; The first interest keyword and the interest renewal rate of the first interest keyword stored in the interest database are deleted, and the interest renewal rates of the remaining interest keywords in the interest database are updated using a preset time migration rate, wherein the first interest keyword is an interest keyword whose feature similarity with the news keyword is greater than a preset similarity threshold, and the updated interest renewal rate is equal to the product of the preset time migration rate and the interest renewal rate before the update; Deleting the second interest keyword and the interest freshness rate of the second interest keyword from the remaining interest keywords in the interest database, wherein the second interest keyword is an interest keyword whose corresponding interest freshness rate is less than a preset freshness rate threshold; For each of the news keywords, an interest freshness rate with a value of 1 is configured for the news keyword, and the news keyword and its interest freshness rate are added to the interest database.

2. The method according to claim 1, characterized in that The step of calculating the current user interest feature vector of the target user according to all keywords included in the interest database includes: Performing word vector conversion on each keyword in the interest database to obtain corresponding keyword vectors; The interest newness rate of each keyword in the interest database is used as a weight, and the obtained keyword vectors are weighted and summed to obtain the user interest feature vector.

3. The method according to claim 1, characterized in that The step of extracting the news word segmentation vector of the news to be recommended includes: Performing word segmentation processing on the news content of the news to be recommended to obtain the word segmentation result of the news to be recommended; The word segmentation result is processed to remove stop words to obtain news word segmentations corresponding to the news to be recommended; The obtained news word segmentation is converted into a word vector to obtain a corresponding news word segmentation vector.

4. The method according to claim 1, characterized in that: The step of calculating the news content feature vector of the news to be recommended based on the extracted news word segmentation vector comprises: Input the news word segmentation vector into the attention mechanism model; The attention mechanism model is called at the word level attention level and the sentence level attention level to perform feature fusion on the news word segmentation vector to obtain the news content feature vector.

5. A news recommendation device, characterized in that: The device comprises: The keyword extraction module is used to extract keywords from the news content currently read by the target user to obtain the corresponding news keywords; A database updating module, configured to adjust the content of the interest keywords stored in the interest database corresponding to the target user according to the news keywords, and add the news keywords to the interest database; A feature vector calculation module, used to calculate the current user interest feature vector of the target user according to all keywords included in the interest database; The feature vector calculation module is further used to extract the news word segmentation vector of the news to be recommended, and calculate the news content feature vector of the news to be recommended based on the extracted news word segmentation vector; A click probability calculation module, used to calculate the click probability of the recommended news being clicked by the target user according to the user interest feature vector and the news content feature vector; A news recommendation module, configured to recommend the news to be recommended to the target user when the click probability is greater than or equal to a preset probability threshold; The database updating module adjusts the content of the interest keywords stored in the interest database corresponding to the target user according to the news keywords, and adds the news keywords to the interest database, including: Calculating feature similarity between each of the news keywords and each of the interest keywords in the interest database; The first interest keyword and the interest renewal rate of the first interest keyword stored in the interest database are deleted, and the interest renewal rates of the remaining interest keywords in the interest database are updated using a preset time migration rate, wherein the first interest keyword is an interest keyword whose feature similarity with the news keyword is greater than a preset similarity threshold, and the updated interest renewal rate is equal to the product of the preset time migration rate and the interest renewal rate before the update; Deleting the second interest keyword and the interest freshness rate of the second interest keyword from the remaining interest keywords in the interest database, wherein the second interest keyword is an interest keyword whose corresponding interest freshness rate is less than a preset freshness rate threshold; For each of the news keywords, an interest freshness rate with a value of 1 is configured for the news keyword, and the news keyword and its interest freshness rate are added to the interest database.

6. The device according to claim 5, characterized in that The feature vector calculation module calculates the current user interest feature vector of the target user according to all keywords included in the interest database, including: Performing word vector conversion on each keyword in the interest database to obtain corresponding keyword vectors; The interest newness rate of each keyword in the interest database is used as a weight, and the obtained keyword vectors are weighted and summed to obtain the user interest feature vector.

7. A media server, characterized in that: The invention comprises a processor and a memory, wherein the memory stores a computer program executable by the processor, and the processor can execute the computer program to implement the news recommendation method described in any one of claims 1 to 4.

8. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the news recommendation method described in any one of claims 1 to 4 is implemented.

Images