Content recommendation method, system, terminal device, medium and product

Abstract

The application relates to the field of recommendation systems and provides a content recommendation method, a content recommendation system, a terminal device, a medium and a product. The method comprises the following steps: after receiving to-be-recommended content, identifying each to-be-recommended content to obtain content information and label information corresponding to each to-be-recommended content, and obtaining user information and historical behavior information corresponding to a target user; generating a recommendation data set according to the user information and the historical behavior information corresponding to the target user and the content information and the label information corresponding to each to-be-recommended content; determining a rule graph corresponding to the target user according to the recommendation data set; determining a click probability of each to-be-recommended content for the target user according to the rule graph, and outputting target recommended content corresponding to the maximum click probability to the target user. The method can improve the accuracy of content recommendation, thereby improving the user experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of recommendation systems, and in particular to a content recommendation method, system, terminal device, medium and product. BACKGROUND

[0002] With the development of Internet technology and the progress of society, various APPs of terminal devices are basically provided with a recommendation system. When a user opens an APP for browsing, the recommendation system can recommend various contents to the user.

[0003] The existing recommendation system mainly uses big data analysis or deep learning to recommend relevant contents to the user, that is, historical browsing data of the user in a period of time is acquired, and big data analysis is performed on the historical browsing data, or deep learning is performed by using a machine learning algorithm, so as to predict the contents that the user is likely to browse. When the user opens the APP again for browsing, the recommendation system will recommend relevant contents to the user according to the prediction. However, the contents recommended by the big data analysis or deep learning may not be accurate, resulting in that the user is not satisfied with the recommended contents.

[0004] Therefore, there is a need for a content recommendation scheme with higher recommendation accuracy. SUMMARY

[0005] The present application provides a content recommendation method, system, terminal device, medium and product to solve the technical problem of low recommendation accuracy of the existing content recommendation system.

[0006] In a first aspect, the present application provides a content recommendation method, comprising:

[0007] After receiving the to-be-recommended contents, identifying each of the to-be-recommended contents to acquire content information and label information corresponding to each of the to-be-recommended contents, and to acquire user information and historical behavior information corresponding to a target user;

[0008] Generating a recommendation data set according to the user information and historical behavior information corresponding to the target user, and the content information and label information corresponding to each of the to-be-recommended contents;

[0009] Determining a rule graph corresponding to the target user according to the recommendation data set;

[0010] Determining a click probability of each of the to-be-recommended contents by the target user according to the rule graph, and outputting a target recommended content corresponding to a maximum click probability to the target user.

[0011] In a possible implementation, the determining of the rule graph corresponding to the target user according to the recommendation data set specifically comprises:

[0012] constructing a recommendation rule corresponding to the historical behavior information of the target user by using a rule learning algorithm, the recommendation rule comprising condition information and result information;

[0013] model fitting is performed on each of the recommendation rules to determine a rule weight corresponding to each of the recommendation rules, the rule weight comprising one or more of recall rate, precision rate and accuracy rate;

[0014] determining a rule graph corresponding to the target user according to user information of the target user, content information of to-be-recommended content, label information corresponding to each of the content information, the recommendation rule corresponding to the target user and the rule weight corresponding to the recommendation rule.

[0015] In a possible implementation, the determining a rule graph corresponding to the target user according to user information of the target user, content information of to-be-recommended content, label information corresponding to each of the content information, the recommendation rule corresponding to the target user and the rule weight corresponding to the recommendation rule specifically comprises:

[0016] constructing a user entity according to the user information of the target user, the user entity comprising a user attribute corresponding to each of the user information;

[0017] constructing a content entity according to the content information of the to-be-recommended content, the content entity comprising a content attribute corresponding to each of the content information;

[0018] constructing a label entity according to the label information corresponding to the content information;

[0019] constructing a rule entity according to the recommendation rule corresponding to the target user and the rule weight corresponding to the recommendation rule;

[0020] constructing a first relationship between the user entity and the rule entity, a second relationship between the content entity and the rule entity and a third relationship between the label entity and the rule entity according to the condition information or the result information of the recommendation rule corresponding to the rule entity;

[0021] constructing a fourth relationship between the content entity and the label entity according to a preset corresponding relationship between the content information and the label information;

[0022] determining the rule graph corresponding to the target user according to the user entity, the content entity, the label entity, the rule entity, the first relationship, the second relationship, the third relationship and the fourth relationship.

[0023] In a possible implementation, the first relationship between the user entity and the rule entity, the second relationship between the content entity and the rule entity, and the third relationship between the label entity and the rule entity are respectively constructed according to the condition information or the result information of the recommended rule corresponding to the rule entity, and specifically include the following steps.

[0024] It is judged whether there is a user attribute in the user entity that satisfies the condition information of the recommended rule corresponding to the rule entity, and if so, a first relationship between the user entity and the rule entity is generated.

[0025] It is judged whether there is a content attribute in the content entity that satisfies the result information of the recommended rule corresponding to the rule entity, and if so, a second relationship between the content entity and the rule entity is generated.

[0026] It is judged whether the label entity satisfies the result information of the recommended rule corresponding to the rule entity, and if so, a third relationship between the label entity and the rule entity is generated.

[0027] In a possible implementation, the click probability of the target user on each of the to-be-recommended contents is determined according to the rule graph, and specifically includes the following steps.

[0028] The user entity vector corresponding to the user entity, the content entity vector corresponding to the content entity, the label entity vector corresponding to the label entity, the rule entity vector corresponding to the rule entity, and the relationship vector corresponding to each relationship in the rule graph are determined, and the rule entity vector is the rule weight corresponding to the recommended rule in the rule entity.

[0029] A first rule entity vector that has a relationship with each user entity vector is obtained, and the user entity vector, the first rule entity vector, and the relationship vector between the user entity vector and the first rule entity vector are spliced to generate a user rule vector.

[0030] A second rule entity vector that has a relationship with each content entity vector or label entity vector is obtained, and the content entity vector or label entity vector, the second rule entity vector, and the relationship vector between the content entity vector or label entity vector and the second rule entity vector are spliced to generate a content rule vector.

[0031] The user rule vector and the content rule vector are randomly spliced to generate a spliced vector.

[0032] The spliced vector is input into a trained neural network model to determine the click probability of the target user on each of the to-be-recommended contents.

[0033] In a possible implementation, the outputting the target recommendation content corresponding to the maximum click probability to the target user specifically includes:

[0034] determining the maximum click probability in the click probabilities, and a target recommendation content corresponding to the maximum click probability;

[0035] determining a recommendation rule corresponding to the maximum click probability, and result information of the recommendation rule;

[0036] generating a recommendation reason corresponding to the target recommendation content according to the result information;

[0037] outputting the target recommendation content and the recommendation reason corresponding to the target recommendation content to the target user.

[0038] In a possible implementation, before the determining the rule graph corresponding to the target user according to the recommendation dataset, the method further includes:

[0039] preprocessing the recommendation dataset to arrange the recommendation dataset into machine learning available data;

[0040] Correspondingly, the determining the rule graph corresponding to the target user according to the recommendation dataset includes:

[0041] determining the rule graph corresponding to the target user according to the recommendation dataset after preprocessing.

[0042] In a second aspect, the present application provides a content recommendation system, which includes:

[0043] a transceiving module, configured to, after receiving to-be-recommended contents, identify each to-be-recommended content to obtain content information and label information corresponding to each to-be-recommended content, and obtain user information and historical behavior information corresponding to a target user;

[0044] a processing module, configured to generate a recommendation dataset according to the user information and historical behavior information corresponding to the target user, and the content information and label information corresponding to each to-be-recommended content; determine a rule graph corresponding to the target user according to the recommendation dataset; determine a click probability of each to-be-recommended content by the target user according to the rule graph; and output a target recommendation content corresponding to a maximum click probability to the target user.

[0045] In a third aspect, the present application provides a terminal device, which includes a processor and a memory in communication connection with the processor.

[0046] The memory stores computer execution instructions.

[0047] The processor executes computer execution instructions stored in the memory to implement the above-described method.

[0048] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the above-described method.

[0049] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0050] The content recommendation method, system, terminal device, medium, and product provided in this application can, after receiving content to be recommended, identify each piece of content to be recommended to obtain content information and tag information corresponding to each piece of content to be recommended, and obtain user information and historical behavior information corresponding to the target user; generate a recommendation dataset based on the user information and historical behavior information corresponding to the target user, and the content information and tag information corresponding to each piece of content to be recommended; determine the rule graph corresponding to the target user based on the recommendation dataset; determine the click probability of the target user for each piece of content to be recommended based on the rule graph, and output the target recommended content corresponding to the maximum click probability to the target user. The method of this application can first generate a rule graph using the recommendation dataset containing the historical behavior information corresponding to the target user, and then construct the rule graph by combining rule learning and knowledge graph, so that the rule graph contains rules that conform to user behavior. Furthermore, after obtaining the rule graph with rules, the click probability of the target user for each piece of content to be recommended in the recommendation dataset can be determined based on the rule graph, and the target recommended content corresponding to the maximum click probability can be output to the target user. This setup ensures that the recommended content corresponding to the highest click probability is predicted by learning the rules corresponding to the target user. This makes the recommendation results interpretable and understandable, improving their accuracy and credibility, attracting more user attention, and enhancing the user experience. Attached Figure Description

[0051] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0052] Figure 1 This is a flowchart of a content recommendation method according to an embodiment of this application;

[0053] Figure 2 A flowchart illustrating a content recommendation method according to another embodiment of this application;

[0054] Figure 3 This is a schematic diagram of a rule map according to an embodiment of this application;

[0055] Figure 4 This is a schematic diagram of the structure of a content recommendation system according to an embodiment of this application;

[0056] Figure 5 This is a schematic diagram of the structure of a terminal device according to an embodiment of this application.

[0057] Reference numerals: 41, transceiver module; 42, processing module.

[0058] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0059] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0060] The collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0061] It should be noted that the methods, systems, terminal devices, media, and products recommended in this application can be used in the financial field, as well as in any other field. The application areas of the methods, systems, terminal devices, media, and products recommended in this application are not limited.

[0062] With the development of internet technology and social progress, most mobile apps on terminal devices have recommendation systems. When users open an app and browse, these systems can recommend various content. For example, a bank's app has an "Information" section. When a user clicks on this section, the app's recommendation system will suggest information including finance and wealth management topics.

[0063] Existing recommendation systems primarily utilize big data analytics or deep learning to recommend relevant content to users. This involves acquiring a user's historical browsing data over a given period and performing big data analysis on this data, or using machine learning algorithms for deep learning to predict the content the user is likely to browse. When the user reopens the app, the recommendation system recommends relevant content based on these predictions to increase the probability of the user clicking on that content.

[0064] However, content recommended through big data analysis or deep learning may not be accurate, leading to user dissatisfaction. Users may feel that the content recommended by the app is not what they need, causing them to lose trust in the app and become unwilling to accept its recommendations.

[0065] Based on this technical problem, the inventive concept of this application is: how to provide a content recommendation method with higher recommendation accuracy.

[0066] The content recommendation method provided in this application aims to solve the aforementioned technical problems of existing technologies. This method first generates a rule graph using a recommendation dataset containing historical behavior information corresponding to the target user. Then, it combines rule learning and knowledge graphs to construct the rule graph, ensuring that the rule graph contains rules consistent with user behavior. Furthermore, after obtaining the rule graph with rules, the click probability of the target user for each piece of content to be recommended in the recommendation dataset can be determined based on the rule graph, and the target recommended content corresponding to the highest click probability is output to the target user. Through this setup, the target recommended content corresponding to the highest click probability is predicted by learning the rules corresponding to the target user, thus making the recommendation results interpretable and understandable, improving the accuracy and credibility of the recommendation results, attracting more user attention, and enhancing the user experience.

[0067] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0068] Example 1

[0069] Figure 1 This is a flowchart of a content recommendation method provided in an embodiment of this application. The execution subject of the content recommendation method provided in this embodiment can be a terminal device or a content recommendation system in the terminal device. This embodiment describes the content recommendation method with the terminal device as the execution subject. Figure 1 As shown, this content recommendation method may include the following steps:

[0070] S101: After receiving the content to be recommended, identify each piece of content to be recommended to obtain the content information and tag information corresponding to each piece of content to be recommended, and obtain the user information and historical behavior information corresponding to the target user.

[0071] In this embodiment, the terminal device includes, but is not limited to, devices such as mobile phones, computers, tablets, and smartwatches, and no restrictions are imposed here.

[0072] In this embodiment, the server can send content to be recommended to the terminal device at regular intervals. After receiving the content to be recommended from the server, the terminal device can identify each piece of content to be recommended in order to obtain the content information and tag information corresponding to each piece of content. For example, the terminal device can use image recognition technology, OCR recognition technology, etc. to identify each piece of content to be recommended. The above-mentioned recognition technologies can all adopt existing recognition technologies, which will not be elaborated here.

[0073] In this embodiment, content information refers to the specific related content of the content to be recommended; tag information refers to the specific aspect of the content to be recommended, such as finance, news, real estate, entertainment, etc. Those skilled in the art can pre-set the correspondence between content information and tag information. Once the terminal device obtains the content information corresponding to the content to be recommended, it can determine the tag information corresponding to the content to be recommended based on the preset correspondence. The specific tag information, and the correspondence between content information and tag information, can be flexibly set by those skilled in the art, and no restrictions are imposed here.

[0074] In this embodiment, user information refers to basic user information such as username, mobile phone number, name, gender, and age; historical behavior information refers to the historical traces retained and recorded when a user performs related operations in the APP on the terminal device, such as browsing history.

[0075] In this embodiment, when a user accesses an app on a terminal device to perform related operations, they typically need to log in using user information such as username and mobile phone number. Furthermore, the operations performed after logging in are retained as historical behavior. Therefore, once the target user logs into the app on the terminal device—that is, once the target user is identified—the corresponding user information and historical behavior information can be determined.

[0076] S102: Generate a recommendation dataset based on the user information and historical behavior information of the target user, as well as the content information and tag information of each piece of content to be recommended.

[0077] S103: Determine the rule graph corresponding to the target user based on the recommendation dataset.

[0078] In this embodiment, for a detailed implementation of step S103 of determining the rule graph corresponding to the target user based on the recommendation dataset, please refer to Embodiment 2.

[0079] In one possible implementation, before determining the rule graph corresponding to the target user based on the recommendation dataset in step S103 above, the method may further include: preprocessing the recommendation dataset to organize it into data usable for machine learning.

[0080] Accordingly, step S103 above, which determines the rule graph corresponding to the target user based on the recommendation dataset, may include: determining the rule graph corresponding to the target user based on the preprocessed recommendation dataset.

[0081] In this embodiment, the recommendation dataset acquired by the terminal device may have issues such as missing data, duplicate data, or inconsistent formatting, making it unusable for machine learning. Therefore, after acquiring the recommendation dataset, it can be preprocessed, such as by performing data imputation, data deletion, or format standardization, to transform the dataset into usable data for machine learning. This approach avoids the inability of the recommendation dataset to perform subsequent machine learning, thereby improving the accuracy of rule graph determination.

[0082] S104: Determine the click probability of the target user for each piece of content to be recommended based on the rule graph, and output the target recommended content corresponding to the maximum click probability to the target user.

[0083] In this embodiment, the click probability of a target user for each piece of content to be recommended can be determined based on the rule graph, and the target recommended content corresponding to the highest click probability is the content that the terminal device should output to the target user.

[0084] In one possible implementation, the step S104 above, which involves determining the click probability of the target user for each piece of content to be recommended based on the rule graph, may include:

[0085] S1041: Determine the user entity vector corresponding to the user entity, the content entity vector corresponding to the content entity, the tag entity vector corresponding to the tag entity, the rule entity vector corresponding to the rule entity, and the relation vector corresponding to each relation in the rule graph. The rule entity vector is the rule weight corresponding to the recommendation rule in the rule entity.

[0086] S1042: Obtain the first rule entity vector that has a relationship with each user entity vector, and concatenate the user entity vector, the first rule entity vector, and the relationship vector between the user entity vector and the first rule entity vector to generate the user rule vector.

[0087] S1043: Obtain the second rule entity vector that has a relationship with each content entity vector or tag entity vector, and concatenate the content entity vector or tag entity vector, the second rule entity vector, and the relationship vector between the content entity vector or tag entity vector and the second rule entity vector to generate the content rule vector.

[0088] S1044: Randomly concatenate the user rule vector and the content rule vector to generate a concatenated vector.

[0089] S1045: Input the concatenated vector into the trained neural network model to determine the probability of the target user clicking on each piece of content to be recommended.

[0090] In this embodiment, the rule graph includes user entities, content entities, tag entities, rule entities, and relationships between any two entities. When calculating vectors, the rule weights corresponding to the recommendation rules within a rule entity can be used as the rule entity vectors corresponding to that rule entity. Furthermore, existing knowledge representation learning algorithms, such as the TransE algorithm, can be used to vectorize the feature graph, calculating the user entity vector corresponding to the user entity, the content entity vector corresponding to the content entity, the tag entity vector corresponding to the tag entity, and the relationship vectors corresponding to each relationship. The feature vectorization of the rule graph using knowledge representation learning algorithms can employ existing calculation methods, which will not be elaborated upon here.

[0091] In this implementation, the user rule vector refers to the concatenation of all vectors related to user information, including user entity vectors, first rule entity vectors related to each user entity vector, and the relationship vector between them. Similarly, the content rule vector refers to the concatenation of all vectors related to content information, including content entity vectors or tag entity vectors, second rule entity vectors related to each content entity vector or tag entity vector, and the relationship vector between them. Since content entities correspond to tag entities, when calculating the content rule vector, concatenation can be based solely on content entity vectors or solely on tag entity vectors.

[0092] In this embodiment, when concatenating user rule vectors and content rule vectors, a many-to-many random concatenation can be performed to randomly generate multiple concatenated vectors, thereby improving the accuracy of click probability. The trained neural network model can be an LSTM neural network model or other neural network models, without limitation. When training the neural network model, the vectors corresponding to the historical behavior information of the target user can be input into the initial neural network model for training to obtain a neural network model capable of classifying recommended content. The classification probability of the recommended content is its corresponding click probability.

[0093] In this implementation, each user rule vector includes all rule entity vectors related to a user entity vector, and each content rule vector also includes all rule entity vectors related to a content entity vector or a tag entity vector. Therefore, after randomly concatenating the user and content rule vectors in a many-to-many manner, the generated concatenated vector includes all possible combinations of user-side and content-side information. After inputting the concatenated vector into a trained neural network model, the model can classify the concatenated vector to obtain the classification probability of the content to be recommended, i.e., the probability of a user clicking on the recommended content. This setup allows for a simple and accurate prediction of the user's click probability on the recommended content, thus improving the accuracy of the click probability prediction.

[0094] In one possible implementation, step S104 above, which involves outputting the target recommended content corresponding to the maximum click probability to the target user, may include: determining the maximum click probability in the click probability and the target recommended content corresponding to the maximum click probability; determining the recommendation rule corresponding to the maximum click probability and the result information of the recommendation rule; generating a recommendation reason corresponding to the target recommended content based on the result information; and outputting the target recommended content and the recommendation reason corresponding to the target recommended content to the target user.

[0095] In this implementation, the recommendation rule is presented as "IF...THEN...", where "THEN..." represents the result information. The result information can be used to generate a recommendation reason corresponding to the target recommended content. For example, if the recommendation rule is "IF user is between 25 and 35 years old THEN user is interested in real estate-related content", the corresponding recommendation reason could be "You have recently been interested in real estate-related content".

[0096] In this embodiment, after determining the maximum click probability and the target recommended content corresponding to the maximum click probability, in order to improve the credibility of the recommendation and make users trust the content recommended by the recommendation system, the recommendation reason corresponding to the target recommended content can also be output at the same time as outputting the target recommended content to the target user, thereby further improving the accuracy and credibility of the recommendation.

[0097] The method in this application first generates a rule graph using a recommendation dataset containing historical behavior information of the target user. This rule graph is then constructed by combining rule learning and knowledge graphs, ensuring it contains rules consistent with user behavior. Furthermore, after obtaining the rule graph with rules, the probability of the target user clicking on each piece of content to be recommended in the recommendation dataset can be determined based on the rule graph. The target recommended content corresponding to the highest click probability is then output to the target user. This setup ensures that the target recommended content corresponding to the highest click probability is predicted by learning the rules corresponding to the target user, thus making the recommendation results interpretable and understandable, improving the accuracy and reliability of the recommendations, attracting more user attention, and enhancing the user experience.

[0098] The following is a detailed description of the specific implementation of step S103 of Example 1 above, which involves determining the rule graph corresponding to the target user based on the recommendation dataset, using Example 2 as an example.

[0099] Example 2

[0100] Figure 2 This is a flowchart of a content recommendation method provided in another embodiment of this application. The execution subject of the content recommendation method provided in this embodiment can be a terminal device or a content recommendation system in the terminal device. This embodiment describes the content recommendation method with the terminal device as the execution subject.

[0101] like Figure 2 As shown, this content recommendation method may include the following steps:

[0102] S201: Construct recommendation rules corresponding to the historical behavior information of the target user using a rule learning algorithm. The recommendation rules include condition information and result information.

[0103] In this embodiment, a rule learning algorithm can be used to construct recommendation rules corresponding to the historical behavior information of target users, thereby learning easily understandable recommendation rules from the historical behavior information of target users. For example, the recommendation rules could be that users aged 25 to 50 pay more attention to financial information; users aged 25 to 35 pay more attention to real estate information; and users who usually pay attention to technology-related content also pay attention to stock market information of technology companies, etc.

[0104] In this embodiment, the recommendation rules can be constructed using existing rule learning algorithms, such as randomized decision trees. A complete rule is formed by starting from the root node and ending at a leaf node. The specific construction method can use existing technologies, which will not be elaborated here. The recommendation rules are represented as "IF……THEN……", where "IF……" corresponds to conditional information and "THEN……" corresponds to result information.

[0105] S202: Perform model fitting for each recommendation rule to determine the rule weight corresponding to each recommendation rule. The rule weight includes one or more of recall, precision, and accuracy.

[0106] In this embodiment, the process of model fitting for the recommendation rules can adopt existing model fitting methods, which will not be elaborated here. After model fitting for the recommendation rules, the recall, precision, or accuracy of the rules can be calculated, and these indicators can be used as the rule weights of the recommendation rules.

[0107] S203: Based on the target user's user information, the content information of the content to be recommended, the tag information corresponding to each content information, the recommendation rules corresponding to the target user, and the rule weights corresponding to the recommendation rules, determine the rule graph corresponding to the target user.

[0108] In this embodiment, a rule graph is a structured semantic knowledge base used to quickly describe concepts and their relationships in the physical world, i.e., "entity-relationship-entity" triples, enabling rapid knowledge response and reasoning. For example, a rule graph KG can be represented as user entity U, content entity I, tag entity L, rule entity R, and the relationships r between entities, i.e., KG{eε(U, I, L, R), rε(r1, r2, r3, r4)}.

[0109] In one possible implementation, step S203 above, which determines the rule graph corresponding to the target user based on the target user's user information, the content information of the content to be recommended, the tag information corresponding to each piece of content information, the recommendation rule corresponding to the target user, and the rule weight corresponding to the recommendation rule, may include:

[0110] S2031: Construct a user entity based on the user information of the target user. The user entity includes user attributes corresponding to each user information.

[0111] S2032: Construct a content entity based on the content information of the content to be recommended. The content entity includes content attributes corresponding to each piece of content information.

[0112] S2033: Construct a tag entity based on the tag information corresponding to the content information.

[0113] S2034: Construct a rule entity based on the recommendation rule corresponding to the target user and the rule weight corresponding to the recommendation rule.

[0114] S2035: Based on the condition information or result information of the recommendation rule corresponding to the rule entity, construct the first relationship between the user entity and the rule entity, the second relationship between the content entity and the rule entity, and the third relationship between the tag entity and the rule entity.

[0115] S2036: Based on the pre-defined correspondence between content information and tag information, construct a fourth relationship between content entities and tag entities.

[0116] S2037: Determine the rule graph corresponding to the target user based on the user entity, content entity, tag entity, rule entity, first relationship, second relationship, third relationship, and fourth relationship.

[0117] In this embodiment, the user attributes correspond one-to-one with the user information in Embodiment 1 above; the content attributes correspond one-to-one with the content information in Embodiment 1 above.

[0118] In this implementation, after obtaining the recommendation rules, a rule-based knowledge graph needs to be built, i.e., a rule graph, which determines the relationships between user entities, content entities, and tag entities and rule entities. For a given rule entity, the relationships between user entities, content entities, and tag entities and rule entities can be easily and accurately determined based on the conditional or result information of the recommendation rule corresponding to that rule entity. Furthermore, since content and tags are corresponding, the relationship between content entities and tag entities can be determined based on the correspondence between content information and tag information. Through this setup, the rule graph corresponding to the target user can be determined simply and accurately.

[0119] In one possible implementation, step S2035, based on the condition information or result information of the recommendation rule corresponding to the rule entity, constructs a first relationship between the user entity and the rule entity, a second relationship between the content entity and the rule entity, and a third relationship between the tag entity and the rule entity. This may include: determining whether there is a user attribute in the user entity that satisfies the condition information of the recommendation rule corresponding to the rule entity; if so, generating a first relationship between the user entity and the rule entity; determining whether there is a content attribute in the content entity that satisfies the result information of the recommendation rule corresponding to the rule entity; if so, generating a second relationship between the content entity and the rule entity; and determining whether the tag entity satisfies the result information of the recommendation rule corresponding to the rule entity; if so, generating a third relationship between the tag entity and the rule entity.

[0120] In this implementation, if a user entity does not contain user attributes that satisfy the condition information of the recommendation rule corresponding to the rule entity, no relationship is generated between the user entity and the rule entity. Similarly, if a content entity does not contain content attributes that satisfy the result information of the recommendation rule, no relationship is generated between the content entity and the rule entity; if a tag entity does not satisfy the result information of the recommendation rule, no relationship is generated between the tag entity and the rule entity.

[0121] In this implementation, the user attributes in the user entity are related to the condition information of the recommendation rule, i.e., whether the user entity meets the conditions of the recommendation rule; while the content attributes and tag entities in the content entity are related to the result information of the recommendation rule, i.e., whether the content entity and tag entity meet the result of the recommendation rule. Therefore, if the user entity has user attributes that meet the condition information of the recommendation rule, a first relationship between the user entity and the rule entity can be generated; if the content entity has content attributes that meet the result information of the recommendation rule, a second relationship between the content entity and the rule entity can be generated; and if the tag entity meets the result information of the recommendation rule, a third relationship between the tag entity and the rule entity can be generated. With this setup, the first relationship between the user entity and the rule entity, the second relationship between the content entity and the rule entity, and the third relationship between the tag entity and the rule entity can be constructed simply and accurately.

[0122] For example, Figure 3 This is a schematic diagram of a rule map according to an embodiment of this application, such as... Figure 3 As shown, U represents the user entity, I represents the content entity, L represents the tag entity, and R represents the rule entity; r1 represents the first relationship between the user entity U and the rule entity R, r2 represents the second relationship between the content entity I and the rule entity R, r3 represents the third relationship between the tag entity L and the rule entity R, and r4 represents the fourth relationship between the content entity I and the tag entity L.

[0123] In this embodiment, a rule learning algorithm can be used to construct recommendation rules corresponding to the target user's historical behavior information, thereby learning easily understandable recommendation rules from the target user's historical behavior information. Then, model fitting can be performed on the recommendation rules to determine the rule weights corresponding to the recommendation rules, thereby improving the accuracy and reliability of the recommendation results. After obtaining the recommendation rules and rule weights, a rule-based knowledge graph, i.e., a rule graph, can be built based on the recommendation rules, user information, content information, and tag information. Through this setup, rule learning and the knowledge graph can be combined to construct a rule graph containing rules that conform to user behavior. This facilitates the subsequent determination of the target user's click probability for each piece of content to be recommended in the recommendation dataset based on the rule graph, thereby further improving the accuracy and reliability of the recommendation results.

[0124] The following specific embodiment illustrates the content recommendation method of this application.

[0125] Example 3

[0126] In one specific embodiment, when a user logs into a bank's app to perform related business operations, the app's content recommendation system will recommend content in the content recommendation section on the app's display interface. The specific content recommendation process is as follows:

[0127] The first step is that after receiving the content to be recommended from the server, the content recommendation system identifies each piece of content to be recommended in order to obtain the content information and tag information corresponding to each piece of content to be recommended. The content recommendation system also obtains the user information and historical behavior information of the target user after the user logs into the bank's APP.

[0128] The second step involves the content recommendation system using rule learning algorithms to construct recommendation rules corresponding to the target user's historical behavior information.

[0129] The third step involves the content recommendation system fitting a model to each recommendation rule in order to determine the rule weight corresponding to each recommendation rule.

[0130] The fourth step involves the content recommendation system determining the rule graph corresponding to the target user based on the user information of the target user, the content information of the content to be recommended, the tag information corresponding to each piece of content, the recommendation rules corresponding to the target user, and the rule weights corresponding to the recommendation rules.

[0131] The fifth step involves the content recommendation system determining the probability of a target user clicking on each piece of content to be recommended based on the rule graph, and identifying the maximum click probability and the target recommended content corresponding to the maximum click probability.

[0132] The sixth step involves the content recommendation system determining the recommendation rule corresponding to the maximum click probability, as well as the result information of the recommendation rule, and generating the recommendation reason corresponding to the target recommended content based on the result information.

[0133] The seventh step is for the content recommendation system to output the target recommended content and the corresponding recommendation reasons to the target user.

[0134] Figure 4 This is a schematic diagram of the structure of a content recommendation system according to an embodiment of this application, as shown below. Figure 4As shown, the content recommendation system includes: a transceiver module 41, used to identify each piece of content to be recommended after receiving it, to obtain content information and tag information corresponding to each piece of content to be recommended, and to obtain user information and historical behavior information corresponding to the target user; and a processing module 42, used to generate a recommendation dataset based on the user information and historical behavior information corresponding to the target user, as well as the content information and tag information corresponding to each piece of content to be recommended; to determine the rule graph corresponding to the target user based on the recommendation dataset; to determine the click probability of the target user for each piece of content to be recommended based on the rule graph, and to output the target recommended content corresponding to the maximum click probability to the target user. In one embodiment, the specific implementation function of the content recommendation system can be described in steps S101-S104 of Embodiment 1 and steps S201-S203 of Embodiment 2, and will not be repeated here.

[0135] Figure 5 This is a schematic diagram of the structure of a terminal device according to an embodiment of this application, as shown below. Figure 5 As shown, the terminal device includes: a processor 101, and a memory 102 communicatively connected to the processor 101; the memory 102 stores computer execution instructions; the processor 101 executes the computer execution instructions stored in the memory 102 to implement the steps of the content recommendation method in the above method embodiments.

[0136] In the aforementioned terminal device, the memory 102 and the processor 101 are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines, such as a bus connection. The memory 102 stores computer execution instructions that implement data access control methods, including at least one software function module that can be stored in the memory 102 in the form of software or firmware. The processor 101 executes various functional applications and data processing by running the software programs and modules stored in the memory 102.

[0137] The memory 102 may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory 102 stores programs, which are executed by the processor 101 upon receiving execution instructions. Furthermore, the software programs and modules within the memory 102 may include an operating system, which may include various software components and / or drivers for managing system tasks (e.g., memory management, storage device control, power management, etc.) and can communicate with various hardware or software components to provide an operating environment for other software components.

[0138] Processor 101 can be an integrated circuit chip with signal processing capabilities. The aforementioned processor 101 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor.

[0139] An embodiment of this application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the steps of the various method embodiments of this application.

[0140] An embodiment of this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the various method embodiments of this application.

[0141] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the appended claims.

[0142] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A content recommendation method, characterized in that, include: After receiving the content to be recommended, each piece of content to be recommended is identified to obtain the content information and tag information corresponding to each piece of content to be recommended, and to obtain the user information and historical behavior information corresponding to the target user. A recommendation dataset is generated based on the user information and historical behavior information corresponding to the target user, as well as the content information and tag information corresponding to each of the recommended contents; Determine the rule graph corresponding to the target user based on the recommendation dataset; The probability of the target user clicking on each piece of content to be recommended is determined based on the rule graph, and the target recommended content corresponding to the maximum click probability is output to the target user. The step of determining the rule graph corresponding to the target user based on the recommendation dataset specifically includes: A recommendation rule is constructed using a rule learning algorithm, corresponding to the historical behavior information of the target user. The recommendation rule includes condition information and result information. For each of the recommendation rules, a model is fitted to determine the rule weight corresponding to each recommendation rule. The rule weight includes one or more of recall, precision, and accuracy. Based on the user information of the target user, a user entity is constructed, and the user entity includes user attributes corresponding to each user information. Based on the content information of the content to be recommended, a content entity is constructed, and the content entity includes content attributes corresponding to each piece of content information; Based on the tag information corresponding to the content information, construct a tag entity; Construct a rule entity based on the recommendation rules corresponding to the target user and the rule weights corresponding to the recommendation rules; Based on the condition information or result information of the recommendation rule corresponding to the rule entity, a first relationship between the user entity and the rule entity, a second relationship between the content entity and the rule entity, and a third relationship between the tag entity and the rule entity are constructed respectively. Based on the preset correspondence between content information and tag information, a fourth relationship is constructed between the content entity and the tag entity; Based on the user entity, content entity, tag entity, rule entity, first relationship, second relationship, third relationship, and fourth relationship, determine the rule graph corresponding to the target user; Determining the click probability of the target user for each piece of content to be recommended based on the rule graph specifically includes: Determine the user entity vector corresponding to the user entity, the content entity vector corresponding to the content entity, the tag entity vector corresponding to the tag entity, the rule entity vector corresponding to the rule entity, and the relation vector corresponding to each relation in the rule graph. The rule entity vector is the rule weight corresponding to the recommendation rule in the rule entity. Obtain the first rule entity vector that has a relationship with each user entity vector, and concatenate the user entity vector, the first rule entity vector, and the relationship vector between the user entity vector and the first rule entity vector to generate the user rule vector; Obtain a second rule entity vector that has a relationship with each content entity vector or tag entity vector, and concatenate the content entity vector or tag entity vector, the second rule entity vector, and the relationship vector between the content entity vector or tag entity vector and the second rule entity vector to generate a content rule vector; The user rule vector and the content rule vector are randomly concatenated to generate a concatenated vector; The concatenated vector is input into the trained neural network model to determine the click probability of the target user for each of the recommended contents.

2. The method according to claim 1, characterized in that, The step of constructing a first relationship between the user entity and the rule entity, a second relationship between the content entity and the rule entity, and a third relationship between the tag entity and the rule entity based on the condition information or result information of the recommendation rule corresponding to the rule entity, specifically includes: Determine whether there are user attributes in the user entity that satisfy the condition information of the recommendation rule corresponding to the rule entity. If so, generate a first relationship between the user entity and the rule entity. Determine whether there is a content attribute in the content entity that satisfies the recommendation rule corresponding to the rule entity. If so, generate a second relationship between the content entity and the rule entity. Determine whether the tag entity satisfies the result information of the recommendation rule corresponding to the rule entity. If so, generate a third relationship between the tag entity and the rule entity.

3. The method according to claim 1, characterized in that, The step of outputting the target recommendation content corresponding to the maximum click probability to the target user specifically includes: Determine the maximum click probability among the click probabilities, and the target recommended content corresponding to the maximum click probability; Determine the recommendation rule corresponding to the maximum click probability, and the result information of the recommendation rule; Based on the results, generate recommendation reasons corresponding to the target recommended content; Output the target recommended content and the corresponding recommendation reason to the target user.

4. The method according to any one of claims 1-2, characterized in that, Before determining the rule graph corresponding to the target user based on the recommendation dataset, the method further includes: The recommendation dataset is preprocessed to organize it into data usable for machine learning; Accordingly, determining the rule graph corresponding to the target user based on the recommendation dataset includes: Based on the preprocessed recommendation dataset, the rule graph corresponding to the target user is determined.

5. A content recommendation system, comprising: The transceiver module is used to identify each piece of content to be recommended after receiving it, so as to obtain the content information and tag information corresponding to each piece of content to be recommended, and to obtain the user information and historical behavior information corresponding to the target user. The processing module is used to generate a recommendation dataset based on the user information and historical behavior information corresponding to the target user, as well as the content information and tag information corresponding to each of the recommended contents; The rule graph corresponding to the target user is determined based on the recommendation dataset; the click probability of the target user for each piece of content to be recommended is determined based on the rule graph, and the target recommended content corresponding to the maximum click probability is output to the target user; The processing module is specifically used to construct recommendation rules corresponding to the historical behavior information of the target user using a rule learning algorithm. The recommendation rules include condition information and result information. For each of the recommendation rules, a model is fitted to determine the rule weight corresponding to each recommendation rule. The rule weight includes one or more of recall, precision, and accuracy. Based on the user information of the target user, a user entity is constructed, and the user entity includes user attributes corresponding to each user information. Based on the content information of the content to be recommended, a content entity is constructed, and the content entity includes content attributes corresponding to each piece of content information; Based on the tag information corresponding to the content information, construct a tag entity; Construct a rule entity based on the recommendation rules corresponding to the target user and the rule weights corresponding to the recommendation rules; Based on the condition information or result information of the recommendation rule corresponding to the rule entity, a first relationship between the user entity and the rule entity, a second relationship between the content entity and the rule entity, and a third relationship between the tag entity and the rule entity are constructed respectively. Based on the preset correspondence between content information and tag information, a fourth relationship is constructed between the content entity and the tag entity; Based on the user entity, content entity, tag entity, rule entity, first relationship, second relationship, third relationship, and fourth relationship, determine the rule graph corresponding to the target user; The processing module is specifically used to determine the user entity vector corresponding to the user entity, the content entity vector corresponding to the content entity, the tag entity vector corresponding to the tag entity, the rule entity vector corresponding to the rule entity, and the relation vector corresponding to each relation in the rule graph. The rule entity vector is the rule weight corresponding to the recommendation rule in the rule entity. Obtain the first rule entity vector that has a relationship with each user entity vector, and concatenate the user entity vector, the first rule entity vector, and the relationship vector between the user entity vector and the first rule entity vector to generate the user rule vector; Obtain a second rule entity vector that has a relationship with each content entity vector or tag entity vector, and concatenate the content entity vector or tag entity vector, the second rule entity vector, and the relationship vector between the content entity vector or tag entity vector and the second rule entity vector to generate a content rule vector; The user rule vector and the content rule vector are randomly concatenated to generate a concatenated vector; The concatenated vector is input into the trained neural network model to determine the click probability of the target user for each of the recommended contents.

6. A terminal device, comprising a processor and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 4.

8. A computer program product comprising a computer program that, when executed by a processor, implements the method of any one of claims 1-4.

Images