Personalized recommendation method, system and device, medium and product

Abstract

The invention discloses a personalized recommendation method, system and device, a medium and a product, and relates to the technical field of big data processing.The social relation strength between users with social relations is calculated by obtaining behavior data of all the users; extracting interest theme distribution of each user from the behavior data, and calculating interest similarity among the users; according to the interest similarity and the social relation strength, calculating the interest propagation strength between the users; according to the interest propagation intensity and the interest theme distribution, calculating an interest score of each user for a specific theme; and generating a recommendation result of each user according to the interest score. By adopting the embodiment of the invention, the interest of the user can be mined and mapped by fusing the social information and the behavior data of the user, the precision of personalized recommendation is improved, and meanwhile, the cold start speed and precision are reduced.

Description

Technical Field

[0001] This invention relates to the field of big data processing technology, and in particular to a personalized recommendation method, system, device, medium, and product. Background Technology

[0002] High-quality personalized recommendations are not only a key to improving user satisfaction, but also an important way for platforms to enhance user stickiness and realize commercial value.

[0003] Current personalized recommendation algorithms are typically either collaborative filtering or content-based. Collaborative filtering analyzes the similarity between users or items, uses historical user behavior data to predict a user's preference for new items, and then generates recommendations. Content-based recommendations analyze the content features of items, build feature representations for those items, and then combine this with the user's historical preferences to generate recommendations.

[0004] However, existing solutions rely too heavily on users' personal historical behavior, resulting in a relatively static portrayal of user interests. This makes it difficult to reflect the dynamic changes in user interests in real time, which affects the accuracy of recommendation results. Summary of the Invention

[0005] The embodiments of the present invention aim to provide a personalized recommendation method, system, device, medium and product to solve the technical problem that the existing technology relies solely on users' personal historical behavior data for personalized recommendations, which leads to limited recommendation accuracy.

[0006] To address the aforementioned technical problems, a first aspect of this invention provides a personalized recommendation method, comprising: Acquire behavioral data from each user and calculate the strength of social relationships among users who have existing social connections; Extract the distribution of interest topics for each user from the behavioral data, and calculate the interest similarity between users; Calculate the interest propagation strength between users based on the interest similarity and the strength of the social relationship; Based on the interest propagation intensity and the interest topic distribution, calculate each user's interest score for a specific topic; Based on the interest scores, recommendation results are generated for each user.

[0007] As an improvement to the above solution, the step of acquiring the behavioral data of each user and calculating the strength of the social relationship between users with existing social relationships includes: Construct a social network graph with users as nodes based on the social relationships between users; Based on the social network graph, obtain the behavioral data of each node and the interaction behavior data of adjacent nodes; Based on the behavioral data and the interaction behavior data, the strength of social relationships between nodes in the social network graph is calculated; The edge weights of the social network graph are obtained based on the strength of the social relationships.

[0008] As an improvement to the above scheme, the step of obtaining behavioral data of each node and interaction behavioral data of adjacent nodes based on the social network graph includes: Obtain the number of communities, comments, and likes for each node in the social network graph as behavioral data for the node; Obtain the number of shared communities, comment interactions, and like interactions of adjacent nodes in the social network graph as data on the interaction behavior of adjacent nodes.

[0009] As an improvement to the above scheme, the step of extracting the interest topic distribution of each user from the behavioral data and calculating the interest similarity between users includes: The behavioral data is analyzed using a pre-trained topic model to obtain the distribution of interest topics for each user; Based on the distribution of interest topics, calculate the interest similarity between users with social relationships.

[0010] As an improvement to the above scheme, the step of calculating the interest similarity between users with social relationships based on the interest topic distribution includes: Based on the distribution of interest topics, the number of interactions between users with social relationships on each interest topic is obtained; Based on the number of interactions on the topic, the cosine similarity algorithm is used to calculate the interest similarity between users.

[0011] As an improvement to the above scheme, the step of calculating the interest propagation strength between users based on the interest similarity and the social relationship strength includes: Based on user behavior data, the interest similarity between users, and the strength of social relationships, calculate each user's social influence score; The intensity of interest propagation among users is calculated based on the social influence score, the interest similarity, and the strength of social relationships.

[0012] As an improvement to the above scheme, the step of calculating each user's social influence score based on user behavior data, the similarity of interests among users, and the strength of social relationships includes: The social influence score of each user is initialized based on the preset initial score; The first contributing factor to the social influence score of the first user is determined based on the strength of the social relationship between the first user and the second user, and the social influence score of the second user; the first user and the second user have a social relationship. Based on the behavioral data of the first user, the second contributing factor to the first user's social influence score is obtained; The third contributing factor to the social influence score of the first user is obtained based on the similarity of interests between the first user and the second user, the strength of social relationship, and the social influence score of the second user. Based on the first contribution factor, the second contribution factor, and the third contribution factor, the social influence score formula for each user is obtained; The social influence score formula is iteratively solved until the preset iteration completion condition is met, and the social influence score of each user is obtained.

[0013] As an improvement to the above scheme, the step of calculating the interest propagation strength among users based on the social influence score, the interest similarity, and the social relationship strength includes: Based on the social relationships between users, a user interest graph is established with users as nodes. Obtain users' social relationship strength, social influence score, and interest similarity, and calculate the first interest propagation factor; Obtain the social relationship strength, social influence score, and interest similarity between the user and all adjacent nodes in the user's interest graph, and calculate the second interest propagation factor; The user's interest propagation intensity is calculated by normalizing the first interest propagation factor using the second interest propagation factor. The edge weights of the user interest graph are obtained based on the interest propagation strength.

[0014] As an improvement to the above scheme, the step of obtaining the user's social relationship strength, social influence score, and interest similarity, and calculating the first interest propagation factor, includes: Obtain users' social influence scores, interest similarity, and social relationship strength; Based on a preset first balance parameter, the strength of the social relationship is logarithmically transformed to obtain the first propagation factor; The social influence score is subjected to a power transformation based on a preset second balance parameter to obtain the second propagation factor; Based on the preset third balance parameter, the interest similarity is exponentially transformed to obtain the third propagation factor; Based on the first propagation factor, the second propagation factor, and the third propagation factor, the first interest propagation factor is obtained.

[0015] As an improvement to the above scheme, the step of calculating each user's interest score for a specific topic based on the interest propagation intensity and the interest topic distribution includes: Based on the distribution of interest topics and user behavior data, calculate the user's overall interest in the topics; Based on the user's social relationships, the user's community can be determined. The community interest influence is obtained based on the intensity of interest dissemination among each user in the community. Based on the overall interest level, the community interest influence, and the interest dissemination intensity, calculate each user's interest score for a specific topic.

[0016] As an improvement to the above solution, the step of calculating the user's overall interest in a topic based on the distribution of interest topics and user behavior data includes: Based on user behavior data, the number of user interaction behaviors in each topic is obtained; Based on the number of interaction behaviors, the topic attractiveness ratio of each topic to the user is obtained; Based on the topic attractiveness ratio and the user's interest topic distribution, calculate the user's overall interest in the topic.

[0017] As an improvement to the above solution, the step of obtaining the topic attractiveness ratio of each topic to the user based on the number of interaction behaviors includes: Obtain preset base weights and preset time decay factors for different behavior types; the behavior types include browsing, liking, commenting, and sharing; Calculate the average number of interactions per user across each behavior type based on the number of interactions described for each behavior type. Calculate the update weight for each behavior type based on the preset base weight, the preset time decay factor, and the average number of interactions. Based on the update weights and the number of user interactions under each behavior type, the topic attractiveness ratio of each topic to the user is calculated.

[0018] As an improvement to the above scheme, the step of obtaining the community interest influence based on the interest propagation intensity of each user in the community includes: The community interest centrality is obtained based on the interest propagation intensity of each user in the social network and the community. Based on the behavioral data of users in the community, the interest similarity between users, and the strength of social relationships, calculate the social influence score of each user in the community; The community interest influence is obtained based on the community interest centrality and the social influence score of each user in the community.

[0019] As an improvement to the above scheme, the step of calculating each user's interest score for a specific topic based on the overall interest level, the community interest influence, and the interest dissemination intensity includes: The community interest weight is obtained based on the dissemination of each user's interest and the influence of the community interest. Based on the distribution of user interest topics, user interest weights are obtained; Based on the community interest weight and the user interest weight, the user's overall interest score is weighted to obtain the user's interest score for a specific topic.

[0020] As an improvement to the above scheme, generating recommendation results for each user based on the interest score includes: Based on users' interest scores on different topics, target communities are selected from the current community set and used as recommendations for users.

[0021] A second aspect of this invention provides a personalized recommendation system, comprising: The social relationship strength calculation module is used to acquire behavioral data of each user and calculate the strength of social relationships between users who have social relationships. The interest similarity calculation module is used to extract the distribution of interest topics of each user from the behavioral data and calculate the interest similarity between users; The interest propagation intensity calculation module is used to calculate the interest propagation intensity between users based on the interest similarity and the social relationship strength. The interest score calculation module is used to calculate the interest score of each user for a specific topic based on the interest propagation intensity and the interest topic distribution. The recommendation result generation module is used to generate recommendation results for each user based on the interest score.

[0022] A third aspect of the present invention provides a personalized recommendation device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the personalized recommendation method as described above.

[0023] A fourth aspect of the present invention provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the personalized recommendation method as described above.

[0024] A fifth aspect of the present invention provides a computer program product, the computer program product including a computer program or computer instructions, wherein when the computer program or computer instructions are executed by a processor, the personalized recommendation method described above is performed.

[0025] Compared with existing technologies, the personalized recommendation method, system, device, medium, and product disclosed in this invention acquires behavioral data of each user and calculates the strength of social relationships among users with existing social connections; extracts the distribution of interest topics for each user from the behavioral data and calculates the interest similarity between users; calculates the interest propagation strength between users based on the interest similarity and the strength of the social relationship; calculates the interest score for each user on a specific topic based on the interest propagation strength and the distribution of interest topics; and generates recommendation results for each user based on the interest scores. By integrating users' social information and behavioral data, this invention can mine and map users' interests, improving the accuracy of personalized recommendations while shortening the cold start time and improving accuracy. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the steps of a personalized recommendation method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a social network graph provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a user interest graph provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a personalized recommendation system provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a personalized recommendation device provided in an embodiment of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] In the description and claims, it should be understood that the terms "first," "second," etc., used in the description and claims are only for the purpose of distinguishing the description of the same technical features, and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated, nor necessarily the order of description or chronological order. The terms are interchangeable where appropriate. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature.

[0029] Providing accurate personalized recommendations to users is an important means to improve user experience and enhance user stickiness. Traditional personalized recommendation algorithms only analyze users' personal historical behavior, which not only easily leads users into information cocoons and damages user experience, but also makes it difficult to accurately adapt to users' dynamic changes in interests, and it is impossible to discover potential interest content in advance to achieve predictive push.

[0030] Based on the above considerations, embodiments of the present invention provide a personalized recommendation method. Please refer to [link to relevant documentation]. Figure 1 In this embodiment, the personalized recommendation method is specifically executed through steps S1 to S5: S1. Obtain behavioral data from each user and calculate the strength of social relationships between users who have social connections.

[0031] It should be noted that, in this embodiment of the invention, the social relationship can be a two-way friend relationship between users, an interaction history, joining the same community, or interacting with the same content.

[0032] S2. Extract the distribution of interest topics for each user from the behavioral data, and calculate the interest similarity between users.

[0033] By accurately depicting the distribution of users' interests, we can identify user groups with similar interests, providing a basis for subsequent interest dissemination analysis and ensuring the matching degree between recommended content and users' core interests.

[0034] S3. Calculate the interest propagation strength between users based on the interest similarity and the social relationship strength.

[0035] By integrating interest similarity and social relationship strength, the degree of mutual influence of interests among users is quantified, breaking through the limitations of traditional recommendations that only focus on individual behavior, capturing the dynamic patterns of interest transmission in social networks, and improving the social relevance of recommendations.

[0036] S4. Calculate the interest score of each user for a specific topic based on the interest propagation intensity and the interest topic distribution.

[0037] By combining the distribution of individual interests and the intensity of interest dissemination among users, the calculated interest score retains the user's core interests while incorporating the interest extension brought about by social interaction, making interest quantification more comprehensive and accurate.

[0038] S5. Based on the interest scores, generate recommendation results for each user.

[0039] On the one hand, the above-mentioned scheme breaks through the limitations of traditional recommendations that rely solely on users' personal historical behavior. By calculating the strength of social relationships, interest similarity, and the intensity of interest propagation, it places users in social networks to characterize their interests, making interest representation more three-dimensional and comprehensive. Furthermore, it makes recommendations based on multi-dimensional integrated interest scores, which can not only accurately match users' current interests, but also introduce potentially interesting content through social connections, reduce the negative impact of information cocoons on user experience, and improve the richness and foresight of recommendations.

[0040] On the other hand, since traditional personalized recommendation algorithms rely on users' personal historical behavior, their ability to handle the cold start problem of new users and new items is limited, making it difficult to provide effective recommendation results in the case of sparse data. However, the embodiments of this invention, by mining social information in user behavior data, can analyze users with whom a user has social relationships even when the user has no relevant historical behavior, in order to characterize the user's interests, and can effectively improve the accuracy of cold start.

[0041] For example, for new users, their social relationships are identified through the groups and friends they have joined, and personalized recommendations are made to the new users based on these social relationships.

[0042] As a preferred implementation, step S1, acquiring behavioral data from each user and calculating the strength of social relationships between users with existing social connections, includes: Construct a social network graph with users as nodes based on the social relationships between users; Based on the social network graph, obtain the behavioral data of each node and the interaction behavior data of adjacent nodes; Based on the behavioral data and the interaction behavior data, the strength of social relationships between nodes in the social network graph is calculated; The edge weights of the social network graph are obtained based on the strength of the social relationships.

[0043] In this embodiment of the invention, users and their social connections are converted into a structured graph model, making abstract social relationships concrete into computable and analyzable mathematical objects, providing a basic framework for the subsequent quantification of social relationship strength. Considering that social relationships have different strengths, user behavior data is converted into social relationship strength based on interaction behavior data, objectively measuring the tightness of social connections between users, realizing the quantification of social relationship strength, and further using social relationship strength to improve the edge weight attribute of the social network graph, so that it can be directly used in graph algorithms.

[0044] For example, based on the social relationships between users, a system can be built as follows: Figure 2 The social network diagram shown There are 6 users in this social relationship graph. Each user is a node in the social network graph, and the strength of the social relationship between users is the weight of the edge connecting the user node.

[0045] It should be noted that in this example, the social network graph is a directed graph, and for simplicity, only unidirectional attributes are marked on the graph. In practical applications, the social network graph can be an undirected graph, meaning the influence relationship between users is equal, or a directed graph with bidirectional attributes, meaning the influence relationship between users is relatively independent.

[0046] Further, preferably, the step of obtaining behavioral data of each node and interaction behavioral data of adjacent nodes based on the social network graph includes: Obtain the number of communities, comments, and likes for each node in the social network graph as behavioral data for the node; Obtain the number of shared communities, comment interactions, and like interactions of adjacent nodes in the social network graph as data on the interaction behavior of adjacent nodes.

[0047] In the above scheme, the interaction behavior data between nodes is divided into multiple dimensions, which not only reduces the processing cost of invalid data, but also makes it easier to use different processing methods for behavior data of different dimensions, so that the calculation results are more in line with the real closeness of social relationships.

[0048] Furthermore, in some preferred embodiments, the user and users The strength of social relationships Represented as: (1); in, For users and users The number of shared communities; For users The number of communities; For users The number of communities; For users and users The number of comments and interactions; For users Number of comments; For users Number of comments; For users and users The number of likes and interactions; For users The number of likes; For users The number of likes; , , These are preset balance parameters used to adjust the importance of different interactive behaviors.

[0049] Compared to existing single-dimensional modeling methods, the Social Network Relationship Strength Calculation Algorithm (SNRSC) described above constructs a multi-dimensional interactive data-driven model and uses mathematical operations such as Dice similarity and Jaccard similarity to reflect the differences in the impact of different social behaviors.

[0050] It should be noted that the above-described application of Dice similarity or Jaccard similarity to different interactive behavior data is one embodiment of the present invention. In practical applications, the specific modeling formula will inevitably differ depending on the classification of interactive behavior data. This is foreseeable to those skilled in the art. In other words, the use of other existing similarity calculation algorithms for different dimensions of data will not affect the beneficial effects produced by the embodiments of the present invention.

[0051] As a preferred implementation, step S2, extracting the interest topic distribution of each user from the behavioral data and calculating the interest similarity between users, includes: The behavioral data is analyzed using a pre-trained topic model to obtain the distribution of interest topics for each user; Based on the distribution of interest topics, calculate the interest similarity between users with social relationships.

[0052] It should be noted that the topic model can represent user behavioral text as an interest distribution vector for preset topics, such as sports topics, food and cooking topics, movie and music topics, and technology and digital topics.

[0053] For example, the topic model is an LDA (Latent Dirichlet Allocation) topic model. LDA is an unsupervised machine learning model primarily used to mine latent topic structures from large-scale text data. In this embodiment of the invention, the text data refers to dynamically posted content, comments, likes, etc., extracted from user behavior data.

[0054] In some preferred embodiments, the user Distribution of Interest Topics Represented as: (2); in, For users obtained through pre-trained topic models On the topic The level of interest.

[0055] For example, in a preferred embodiment, there are five interest topics: travel, food, technology, automobiles, and parenting. Each interest topic is coded, with the travel topic represented as follows: Food theme is represented as The technology theme is represented as The car theme is represented as The theme of childcare is represented as For users After analyzing the behavioral data, we obtained the distribution of their interest topics. The distribution of these interest topics reveals that users... People showed the highest interest in travel-themed topics and the lowest interest in parenting-themed topics.

[0056] More preferably, the interest topic distributions of all users on the platform can be combined to obtain an interest distribution matrix by comprehensively analyzing the interest performance of all users across all topics. For example, the interest distribution matrix... In the interest distribution matrix, each row corresponds to a user, and each column corresponds to a topic.

[0057] Further, preferably, the step of calculating the interest similarity between users with social relationships based on the interest topic distribution includes: Based on the distribution of interest topics, the number of interactions between users with social relationships on each interest topic is obtained; Based on the number of interactions on the topic, the cosine similarity algorithm is used to calculate the interest similarity between users.

[0058] By analyzing the distribution of interest topics and the number of topic interactions, we can transform the fuzzy similarity of users' interests into quantitative indicators. Calculating interest similarity using the cosine similarity method is suitable for batch similarity calculations on large-scale user social networks and can effectively measure the degree of fit between interest preferences.

[0059] In some preferred embodiments, the user and users Interest similarity between Represented as: (3); in, Total number of topics; For users On the topic Number of topic interactions; For users On the topic Number of topic interactions.

[0060] In equation (3) above, the user is constructed based on the number of topic interactions. Theme interaction vectors and users Theme interaction vectors Furthermore, based on the L2 norm of each vector and the dot product of two vectors, the interest similarity between users is calculated, thus realizing the calculation of interest similarity between users using the cosine similarity algorithm.

[0061] As a preferred implementation, step S3, calculating the interest propagation strength between users based on the interest similarity and the social relationship strength, is executed through steps S31-S32: S31. Calculate the social influence score for each user based on user behavior data, the interest similarity between users, and the strength of social relationships; S32. Calculate the interest propagation intensity between users based on the social influence score, the interest similarity, and the social relationship strength.

[0062] By combining user behavior data, the similarity of interests among users, and the strength of social relationships, a user's social influence score can be calculated. This score can comprehensively reflect a user's actual ability to promote the spread of interests in social networks. Further calculation of interest spread ability can directly quantify the efficiency of information transmission between different objects, which is conducive to early warning of potential interest topic diffusion trends.

[0063] Further, preferably, step S31, calculating each user's social influence score based on user behavior data, the similarity of interests among users, and the strength of social relationships, includes: The social influence score of each user is initialized based on the preset initial score; The first contributing factor to the social influence score of the first user is determined based on the strength of the social relationship between the first user and the second user, and the social influence score of the second user; the first user and the second user have a social relationship. Based on the behavioral data of the first user, the second contributing factor to the first user's social influence score is obtained; The third contributing factor to the social influence score of the first user is obtained based on the similarity of interests between the first user and the second user, the strength of social relationship, and the social influence score of the second user. Based on the first contribution factor, the second contribution factor, and the third contribution factor, the social influence score formula for each user is obtained; The social influence score formula is iteratively solved until the preset iteration completion condition is met, and the social influence score of each user is obtained.

[0064] It should be noted that the second user is not a specific single user, but refers to all users who have social relationships with the first user; the second user can be regarded as a set of users. Furthermore, the social influence scores of the first user and the second user are interdependent, and are solved using an iterative method.

[0065] In the above scheme, the first contribution factor adopts the PageRank concept, which believes that friends with high influence contribute greater influence; the second contribution factor analyzes the contribution made by the user's own interactive behavior, for example, the more interactive behaviors, the greater the influence; the third contribution factor is obtained through the interest similarity with social friends, that is, neighbors with similar interests contribute greater influence.

[0066] In some preferred embodiments, the first user The formula for social influence score is expressed as: (4); in, To the first user A collection of second users with existing social relationships; For the second user A set of users with social relationships; For the second user Social influence score; For the first user Second User The strength of social relationships can be obtained through equation (1); For the second user The sum of the strength of social relationships between it and all its social users; For the first user The number of interactive behaviors; For the second user The number of interactive behaviors; For the first user Second User The interest similarity can be obtained through equation (3); , , These are the preset balance parameters.

[0067] Furthermore, after obtaining the social influence score formula for each user, the social influence dependency relationship among all users is iteratively solved based on the initial score. In equation (4), for the first user... When solving for social influence, the formula... The results from the previous iteration are used. Through iterative calculation, each user's influence score on the social network can be obtained and used for subsequent social recommendation and influence analysis tasks.

[0068] Preferably, the iterative solution of the social influence score formula is implemented using the power method or the PageRank algorithm. The preset iteration completion condition is that the social influence scores of all users converge or the preset number of iterations is reached.

[0069] Preferably, step S32, calculating the interest propagation intensity among users based on the social influence score, the interest similarity, and the social relationship strength, includes: Based on the social relationships between users, a user interest graph is established with users as nodes. Obtain users' social relationship strength, social influence score, and interest similarity, and calculate the first interest propagation factor; Obtain the social relationship strength, social influence score, and interest similarity between the user and all adjacent nodes in the user's interest graph, and calculate the second interest propagation factor; The user's interest propagation intensity is calculated by normalizing the first interest propagation factor using the second interest propagation factor. The edge weights of the user interest graph are obtained based on the interest propagation strength.

[0070] In some preferred embodiments, the user interest graph is based on a social network graph. The constructed result is represented as ,in, For a set of user nodes, Let be the set of edges for interest propagation among users, and let the edge weights of the user interest graph be the interest propagation strength.

[0071] Furthermore, preferably, the step of obtaining the user's social relationship strength, social influence score, and interest similarity, and calculating the first interest propagation factor, includes: Obtain users' social influence scores, interest similarity, and social relationship strength; Based on a preset first balance parameter, the strength of the social relationship is logarithmically transformed to obtain the first propagation factor; The social influence score is subjected to a power transformation based on a preset second balance parameter to obtain the second propagation factor; Based on the preset third balance parameter, the interest similarity is exponentially transformed to obtain the third propagation factor; Based on the first propagation factor, the second propagation factor, and the third propagation factor, the first interest propagation factor is obtained.

[0072] In some preferred embodiments, the user and users The first interest propagation factor between them is represented as: (5); in, This is the first equilibrium parameter; For users and users The strength of the social relationship between them can be obtained through equation (1); This is the second equilibrium parameter; For users The social influence score can be obtained through equation (4); This is the third equilibrium parameter; For users and users The interest similarity between them can be obtained through equation (3); This is the preset fourth balance parameter.

[0073] In a preferred embodiment of the present invention, each balance parameter is optimized using a cross-validation method.

[0074] Furthermore, corresponding to the first interest propagation factor, users and users The second interest propagation factor between them is expressed as: (6); in, To the first user A set of users who have social relationships.

[0075] Furthermore, users and users Intensity of interest transmission between Represented as: (7).

[0076] In the above scheme, the interest propagation intensity comprehensively considers factors such as the strength of social relationships, social influence, and similarity of user interests. It can more stably, effectively, and robustly model the interest propagation process between users, so as to construct a user interest graph that is more in line with the actual situation.

[0077] In a preferred implementation, step S4, calculating each user's interest score for a specific topic based on the interest propagation intensity and the interest topic distribution, is executed via steps S41-S44: S41. Calculate the user's overall interest in the topics based on the distribution of interest topics and the user's behavioral data; S42. Based on the user's social relationships, obtain the user's community; S43. Based on the intensity of interest propagation among each user in the community, the community interest influence is obtained; S44. Calculate each user's interest score for a specific topic based on the overall interest level, the community interest influence, and the interest dissemination intensity.

[0078] It should be noted that the specific topic is any topic in a preset set of topics, and personalized recommendations can be made based on the matching degree between the content / item and the specific topic.

[0079] The above solution considers both the interests reflected in users' personal behavior and the precise positioning of users' interest-based social circles through social relationships. It captures the shaping effect of the social environment on interests by calculating the community interest influence based on the intensity of interest dissemination among users within the community. The resulting interest score is both based on users' own behavior and can dynamically respond to the influence of the social environment, thus alleviating the problem of information cocoons to some extent.

[0080] Further, preferably, step S41, calculating the user's overall interest in a topic based on the distribution of interest topics and the user's behavioral data, includes: Based on user behavior data, the number of user interaction behaviors in each topic is obtained; Based on the number of interaction behaviors, the user's topic attraction ratio for each topic is obtained; Based on the topic attractiveness ratio and the user's interest topic distribution, calculate the user's overall interest in the topic.

[0081] It should be noted that the topic attractiveness ratio is obtained by normalizing the number of user interactions on a particular topic, which solves the problem that the total number of interactions between different users / topics is different and difficult to compare directly.

[0082] Furthermore, as a preferred implementation, obtaining the user's topic attractiveness ratio for each topic based on the number of interaction behaviors includes: Obtain preset base weights and preset time decay factors for different behavior types; the behavior types include browsing, liking, commenting, and sharing; Calculate the average number of interactions per user across each behavior type based on the number of interactions described for each behavior type. Calculate the update weight for each behavior type based on the preset base weight, the preset time decay factor, and the average number of interactions. Based on the update weights and the number of user interactions under each behavior type, calculate the user's topic attractiveness ratio for each topic.

[0083] It should be noted that different behavior types are assigned initial differentiated weights by preset base weights to reflect the varying degrees of expression of user interests by different behavior types. User interests are time-sensitive; calculating attractiveness solely based on historical behavior accumulation while ignoring the time factor may lead to misjudgments of current user interests. The time decay factor considers the dynamic nature of user behavior over time, dynamically mitigating the impact of outdated behaviors and making the topic attractiveness ratio closer to the user's current interest state, thus improving timeliness and dynamic adaptability. Furthermore, considering the individual differences in behavioral preferences among users, applying uniform behavior weights to all users may not be suitable for these individual habits. Updating the weights allows the weight settings to be dynamically calibrated according to user behavior habits.

[0084] The update weights for each behavior type are updated over time steps. In some preferred embodiments, the behavior type... At time step Update weight Represented as: (8); in, For behavior type The preset base weights; For behavior type At time step The time decay factor is expressed as , It is a type of behavior The time decay rate, This is the current time step; For time steps At that time, all users in terms of behavior type Average number of interactions on; For behavior type At time step noise factor; It is the set of all behavior types.

[0085] Furthermore, all users in terms of behavior type Average number of interactions Represented as: (9); in, Total number of users For users The total number of interactive topics, For time steps At that time, the user On the topic In behavior type The number of interactions on the device.

[0086] Preferably, the noise factor is a random noise term, obtained by random sampling from a normal distribution.

[0087] In some preferred embodiments, the user On the topic The theme attractiveness ratio is expressed as: (10); in, For users A set of behavior types To update the weights, For users On the topic In behavior type Number of interactions on the platform For users For all topics in behavioral type The maximum number of interactive actions on the platform. For any topic in the set of all topics.

[0088] It is understandable that the exponential transformation in equation (10) can amplify the difference in the number of interactions, so that topics with a high number of interactions receive higher weights.

[0089] Furthermore, users On the topic Overall interest Represented as: (11); in, For users On the topic The degree of interest is obtained through the distribution of the user's interest topics.

[0090] The above scheme for calculating overall interest incorporates the degree of interest in the topic. Capture users' personalized interests and preferences on different topics, and update the weights accordingly. Considering the differentiated impact of different behaviors on user interests; using Softmax normalization. The intensity of user behavior on different topics is converted into relative preference probabilities; numerical stabilization is performed using the Log-Sum-Exp Trick to avoid overflow issues; and summation operations are used... By integrating preference information from different behavioral types, a comprehensive expression of user interest is obtained; finally, this is compared with the degree of interest in the topic. Multiply to get the user Overall interest in the topic Compared with traditional recommendation algorithms, the embodiments of the present invention can more accurately, comprehensively, and personally characterize user interests. While alleviating the problems of data sparsity and cold start, it provides users with recommendation results that are more in line with their interests and preferences, which is of great significance for improving user experience and satisfaction.

[0091] Preferably, step S43, obtaining the community interest influence based on the interest propagation intensity of each user in the community, includes: The community interest centrality is obtained based on the interest propagation intensity of each user in the social network and the community. Based on the behavioral data of users in the community, the interest similarity between users, and the strength of social relationships, calculate the social influence score of each user in the community; The community interest influence is obtained based on the community interest centrality and the social influence score of each user in the community.

[0092] By combining the intensity of interest dissemination among users within social networks and communities, the interest influence of the community within the entire social network can be reflected, thus fully considering the impact of the community on user interests and obtaining a more comprehensive interest score.

[0093] In some preferred embodiments, community influence analysis is performed based on the community provided by the platform; in other preferred embodiments, please refer to [link to preferred embodiments]. Figure 3 By combining user interest graphs, we can analyze the influence of community interests and obtain the community based on the connected subgraphs in the user interest graph.

[0094] The innovative Social Network Graph-based Interest Mapping Algorithm (SNGIM) proposed in this invention fully utilizes the social relationship chains and social influence information contained in social networks. Through graph mining technology, it achieves accurate mining and mapping of user interests, solves the limitations of existing algorithms, and provides strong support for personalized recommendations.

[0095] Preferably, community Community interest influence Represented as: (12); in, For users The social influence score can be obtained through equation (4); For users and users The intensity of interest propagation between individuals can be obtained through equation (7); It is the set of nodes in the user interest graph.

[0096] Preferably, step S44, calculating each user's interest score for a specific topic based on the overall interest level, the community interest influence, and the interest dissemination intensity, includes: The community interest weight is obtained based on the dissemination of each user's interest and the influence of the community interest. Based on the distribution of user interest topics, user interest weights are obtained; Based on the community interest weight and the user interest weight, the user's overall interest score is weighted to obtain the user's interest score for a specific topic.

[0097] In some preferred embodiments, the user Interest score Represented as: (13); in, Total number of topics; For users On the topic The overall interest level can be obtained through equation (11); For users On the topic The degree of interest can be obtained through the distribution of interest topics; For community gathering; For the community The influence of community interests can be obtained through formula (12); For the community Chinese users and users The intensity of interest propagation between individuals can be obtained through equation (7); A set of nodes for the user's interest graph.

[0098] Unlike traditional recommendation algorithms based on user-content rating matrices, this approach takes a social network perspective, constructing user interest graphs and group structures to fully explore the social characteristics and collective intelligence of user interests.

[0099] This invention innovatively introduces the indicator of community interest influence on top of individual interests, effectively characterizing a user's interest influence and dissemination ability within social networks. By using a weighted summation method, the relative importance of individual behavior and social influence is flexibly controlled, enabling dynamic expression and real-time updates of user interests. This multi-dimensional, socialized user modeling method provides a new perspective and approach for solving problems such as cold start and information cocoons in recommendation systems.

[0100] As a preferred implementation, step S5, generating recommendation results for each user based on the interest score, includes: Based on users' interest scores on different topics, target communities are selected from the current community set and used as recommendations for users.

[0101] The personalized recommendations given in the above scheme are based on community recommendations. In practical applications, recommendations can also be based on items, content, or information. The specific recommended data does not affect the beneficial effects of the embodiments of the present invention. The embodiments of the present invention are merely a preferred example.

[0102] In a specific application scenario, the platform has multiple interest groups covering various topics such as technological innovation, travel, and lifestyle knowledge. The goal of personalized recommendations is to suggest the most suitable interest groups for each user and conduct targeted marketing campaigns.

[0103] First, a user interest graph is constructed based on data such as users' social relationships and interactive behaviors. In the user interest graph, each node represents a user, and the attributes of the node include the user's basic information and comprehensive interest degree. Each edge represents the interest propagation relationship between two users, and the edge weight is the interest propagation strength.

[0104] Next, based on the user interest graph, an interest score is calculated for each user. Based on the interest score, which reflects the user's preference for different circle topics, circle groups are recommended to the user or targeted marketing is carried out.

[0105] When recommending circles and groups, specifically, it is based on the user. Interest score Find users The platform identifies the most interesting topics, such as travel and technological innovation, and then finds high-quality communities related to these topics to form a candidate recommendation set. Each community group in the candidate recommendation set is then... Calculate users Its interest matching degree ,in, It is a vector of user interest scores. For circles and groups The topic vector. Based on interest matching degree. The candidate recommendation set is sorted, and the top K circles / groups with the highest matching degree are selected as the final recommendation results, which are then displayed to the user. It also records user feedback behavior for subsequent model optimization and evaluation.

[0106] Similarly, in targeted marketing, one can tailor marketing strategies based on user... Interest score Find users Design marketing campaigns around the topics that interest users most, such as launching new communities, etc. When users visit or interact with the platform, these marketing campaigns are shown to them to attract their attention, and user feedback is recorded to evaluate marketing effectiveness and optimize marketing strategies.

[0107] The personalized recommendation method provided in this invention overcomes the limitations of traditional recommendations that rely solely on users' historical behavior. By integrating the strength of social relationships, interest similarity, and the intensity of interest propagation, it constructs a comprehensive and multi-dimensional interest representation. Based on multi-dimensional interest scores, it achieves the dual effect of accurately matching current interests and introducing potential content, reducing the negative impact of information cocoons on user experience and enhancing the richness and foresight of recommendations. Simultaneously, by mining social information, it effectively alleviates the cold start problem for new users and new items. When users lack historical behavior, it relies on social association analysis of interests, significantly improving recommendation accuracy in data-sparse scenarios, ultimately achieving comprehensive optimization of recommendation quality and applicability.

[0108] This invention provides a personalized recommendation system. Please refer to [link / reference]. Figure 4 The personalized recommendation system includes a social relationship strength calculation module 11, an interest similarity calculation module 12, an interest propagation strength calculation module 13, an interest score calculation module 14, and a recommendation result generation module 15, wherein: The social relationship strength calculation module 11 is used to acquire the behavioral data of each user and calculate the social relationship strength between users who have social relationships. The interest similarity calculation module 12 is used to extract the interest topic distribution of each user from the behavioral data and calculate the interest similarity between users; The interest propagation intensity calculation module 13 is used to calculate the interest propagation intensity between users based on the interest similarity and the social relationship strength. The interest score calculation module 14 is used to calculate the interest score of each user for a specific topic based on the interest propagation intensity and the interest topic distribution. The recommendation result generation module 15 is used to generate recommendation results for each user based on the interest score.

[0109] In a preferred embodiment, the social relationship strength calculation module 11 includes: The social network graph construction unit is used to construct a social network graph with users as nodes based on the social relationships between users. The behavior data acquisition unit is used to acquire the behavior data of each node based on the social network graph, and to obtain the interaction behavior data of adjacent nodes. A social relationship strength calculation unit is used to calculate the social relationship strength between nodes in the social network graph based on the behavioral data and the interaction behavior data. The edge weighting unit is used to obtain the edge weights of the social network graph based on the strength of the social relationship.

[0110] Further, preferably, the behavior data acquisition unit is specifically used for: Obtain the number of communities, comments, and likes for each node in the social network graph as behavioral data for the node; Obtain the number of shared communities, comment interactions, and like interactions of adjacent nodes in the social network graph as data on the interaction behavior of adjacent nodes.

[0111] In a preferred embodiment, the interest similarity calculation module 12 includes: The interest topic distribution calculation unit is used to analyze the behavioral data using a pre-trained topic model to obtain the interest topic distribution of each user. The interest similarity calculation unit is used to calculate the interest similarity between users who have social relationships based on the interest topic distribution.

[0112] Further, preferably, the interest similarity calculation unit is specifically used for: Based on the distribution of interest topics, the number of interactions between users with social relationships on each interest topic is obtained; Based on the number of interactions on the topic, the cosine similarity algorithm is used to calculate the interest similarity between users.

[0113] In a preferred embodiment, the interest propagation intensity calculation module 13 includes: The social influence score calculation unit is used to calculate the social influence score of each user based on the user's behavioral data, the interest similarity between users, and the strength of the social relationship. The interest propagation intensity calculation unit is used to calculate the interest propagation intensity between users based on the social influence score, the interest similarity, and the social relationship strength.

[0114] Further, preferably, the social influence score calculation unit is specifically used for: The social influence score of each user is initialized based on the preset initial score; The first contributing factor to the social influence score of the first user is determined based on the strength of the social relationship between the first user and the second user, and the social influence score of the second user; the first user and the second user have a social relationship. Based on the behavioral data of the first user, the second contributing factor to the first user's social influence score is obtained; The third contributing factor to the social influence score of the first user is obtained based on the similarity of interests between the first user and the second user, the strength of social relationship, and the social influence score of the second user. Based on the first contribution factor, the second contribution factor, and the third contribution factor, the social influence score formula for each user is obtained; The social influence score formula is iteratively solved until the preset iteration completion condition is met, and the social influence score of each user is obtained.

[0115] Preferably, the interest propagation intensity calculation unit is specifically used for: Based on the social relationships between users, a user interest graph is established with users as nodes. Obtain users' social relationship strength, social influence score, and interest similarity, and calculate the first interest propagation factor; Obtain the social relationship strength, social influence score, and interest similarity between the user and all adjacent nodes in the user's interest graph, and calculate the second interest propagation factor; The user's interest propagation intensity is calculated by normalizing the first interest propagation factor using the second interest propagation factor. The edge weights of the user interest graph are obtained based on the interest propagation strength.

[0116] Furthermore, preferably, the step of obtaining the user's social relationship strength, social influence score, and interest similarity, and calculating the first interest propagation factor, includes: Obtain users' social influence scores, interest similarity, and social relationship strength; Based on a preset first balance parameter, the strength of the social relationship is logarithmically transformed to obtain the first propagation factor; The social influence score is subjected to a power transformation based on a preset second balance parameter to obtain the second propagation factor; Based on the preset third balance parameter, the interest similarity is exponentially transformed to obtain the third propagation factor; Based on the first propagation factor, the second propagation factor, and the third propagation factor, the first interest propagation factor is obtained.

[0117] In a preferred embodiment, the interest score calculation module 14 includes: The comprehensive interest calculation unit is used to calculate the user's comprehensive interest in the topics based on the distribution of interest topics and the user's behavioral data. The community identification unit is used to determine the community a user belongs to based on their social relationships. The community interest influence calculation unit is used to obtain the community interest influence based on the interest dissemination intensity of each user in the community. The interest score calculation unit is used to calculate each user's interest score for a specific topic based on the overall interest level, the community interest influence, and the interest dissemination intensity.

[0118] Further, preferably, the comprehensive interest calculation unit is specifically used for: Based on user behavior data, the number of user interaction behaviors in each topic is obtained; Based on the number of interaction behaviors, the topic attractiveness ratio of each topic to the user is obtained; Based on the topic attractiveness ratio and the user's interest topic distribution, calculate the user's overall interest in the topic.

[0119] Furthermore, as a preferred implementation, obtaining the topic attractiveness ratio of each topic to the user based on the number of interaction behaviors includes: Obtain preset base weights and preset time decay factors for different behavior types; the behavior types include browsing, liking, commenting, and sharing; Calculate the average number of interactions per user across each behavior type based on the number of interactions described for each behavior type. Calculate the update weight for each behavior type based on the preset base weight, the preset time decay factor, and the average number of interactions. Based on the update weights and the number of user interactions under each behavior type, the topic attractiveness ratio of each topic to the user is calculated.

[0120] Preferably, the community interest influence calculation unit is specifically used for: The community interest centrality is obtained based on the interest propagation intensity of each user in the social network and the community. Based on the behavioral data of users in the community, the interest similarity between users, and the strength of social relationships, calculate the social influence score of each user in the community; The community interest influence is obtained based on the community interest centrality and the social influence score of each user in the community.

[0121] Preferably, the interest score calculation unit is specifically used for: The community interest weight is obtained based on the dissemination of each user's interest and the influence of the community interest. Based on the distribution of user interest topics, user interest weights are obtained; Based on the community interest weight and the user interest weight, the user's overall interest score is weighted to obtain the user's interest score for a specific topic.

[0122] In a preferred embodiment, the recommendation result generation module 15 is specifically used for: Based on users' interest scores on different topics, target communities are selected from the current community set and used as recommendations for users.

[0123] The personalized recommendation system provided by this invention overcomes the limitations of traditional recommendations that rely solely on users' historical behavior. By integrating the strength of social relationships, interest similarity, and the intensity of interest propagation, it constructs a comprehensive and multi-dimensional interest representation. Based on multi-dimensional interest scores, it achieves the dual effect of accurately matching current interests and introducing potential content, reducing the negative impact of information cocoons on user experience and enhancing the richness and foresight of recommendations. Simultaneously, by mining social information, it effectively alleviates the cold start problem for new users and new items. When users lack historical behavior, it relies on social association analysis of interests, significantly improving recommendation accuracy in data-sparse scenarios, ultimately achieving comprehensive optimization of recommendation quality and applicability.

[0124] See Figure 5 , Figure 5 This is a structural block diagram of a personalized recommendation device provided in an embodiment of the present invention. The personalized recommendation device includes a processor 31, a memory 32, and a computer program stored in the memory 32 and executable on the processor 31. When the processor 31 executes the computer program, it implements the steps in the above-described embodiments of the personalized recommendation methods, such as steps S1 to S5.

[0125] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 32 and executed by the processor 31 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the personalized recommendation device.

[0126] The personalized recommendation device may include, but is not limited to, a processor 31 and a memory 32. Those skilled in the art will understand that the schematic diagram is merely an example of a personalized recommendation device and does not constitute a limitation on the device. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, the personalized recommendation device may also include input / output devices, network access devices, buses, etc.

[0127] The processor 31 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 31 is the control center of the personalized recommendation device, connecting various parts of the personalized recommendation device through various interfaces and lines.

[0128] The memory 32 can be used to store the computer programs and / or modules. The processor 31 implements various functions of the personalized recommendation device by running or executing the computer programs and / or modules stored in the memory 32 and calling the data stored in the memory 32. The memory 32 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 32 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0129] If the modules / units integrated into the personalized recommendation device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by the processor 31, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0130] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A personalized recommendation method, characterized in that, include: Acquire behavioral data from each user and calculate the strength of social relationships among users who have existing social connections; Extract the distribution of interest topics for each user from the behavioral data, and calculate the interest similarity between users; Calculate the interest propagation strength between users based on the interest similarity and the strength of the social relationship; Based on the interest propagation intensity and the interest topic distribution, calculate each user's interest score for a specific topic; Based on the interest scores, recommendation results are generated for each user.

2. The personalized recommendation method as described in claim 1, characterized in that, The step of acquiring behavioral data from each user and calculating the strength of social relationships among users with existing social connections includes: Construct a social network graph with users as nodes based on the social relationships between users; Based on the social network graph, obtain the behavioral data of each node and the interaction behavior data of adjacent nodes; Based on the behavioral data and the interaction behavior data, the strength of social relationships between nodes in the social network graph is calculated; The edge weights of the social network graph are obtained based on the strength of the social relationships.

3. The personalized recommendation method as described in claim 2, characterized in that, The step of obtaining behavioral data of each node and interaction behavioral data of adjacent nodes based on the social network graph includes: Obtain the number of communities, comments, and likes for each node in the social network graph as behavioral data for the node; Obtain the number of shared communities, comment interactions, and like interactions of adjacent nodes in the social network graph as data on the interaction behavior of adjacent nodes.

4. The personalized recommendation method as described in claim 1, characterized in that, The step of extracting the interest topic distribution of each user from the behavioral data and calculating the interest similarity between users includes: The behavioral data is analyzed using a pre-trained topic model to obtain the distribution of interest topics for each user; Based on the distribution of interest topics, calculate the interest similarity between users with social relationships.

5. The personalized recommendation method as described in claim 4, characterized in that, The step of calculating the interest similarity between users with social relationships based on the interest topic distribution includes: Based on the distribution of interest topics, the number of interactions between users with social relationships on each interest topic is obtained; Based on the number of interactions on the topic, the cosine similarity algorithm is used to calculate the interest similarity between users.

6. The personalized recommendation method as described in claim 1, characterized in that, The step of calculating the interest propagation strength among users based on the interest similarity and the strength of the social relationship includes: Based on user behavior data, the interest similarity between users, and the strength of social relationships, calculate each user's social influence score; The intensity of interest propagation among users is calculated based on the social influence score, the interest similarity, and the strength of social relationships.

7. The personalized recommendation method as described in claim 6, characterized in that, The calculation of each user's social influence score based on user behavior data, the similarity of interests among users, and the strength of social relationships includes: The social influence score of each user is initialized based on the preset initial score; The first contributing factor to the social influence score of the first user is determined based on the strength of the social relationship between the first user and the second user, and the social influence score of the second user; the first user and the second user have a social relationship. Based on the behavioral data of the first user, the second contributing factor to the first user's social influence score is obtained; The third contributing factor to the social influence score of the first user is obtained based on the similarity of interests between the first user and the second user, the strength of social relationship, and the social influence score of the second user. Based on the first contribution factor, the second contribution factor, and the third contribution factor, the social influence score formula for each user is obtained; The social influence score formula is iteratively solved until the preset iteration completion condition is met, and the social influence score of each user is obtained.

8. A personalized recommendation method as described in claim 6 or 7, characterized in that, The step of calculating the intensity of interest propagation among users based on the social influence score, the interest similarity, and the strength of the social relationship includes: Based on the social relationships between users, a user interest graph is established with users as nodes. Obtain users' social relationship strength, social influence score, and interest similarity, and calculate the first interest propagation factor; Obtain the social relationship strength, social influence score, and interest similarity between the user and all adjacent nodes in the user's interest graph, and calculate the second interest propagation factor; The user's interest propagation intensity is calculated by normalizing the first interest propagation factor using the second interest propagation factor. The edge weights of the user interest graph are obtained based on the interest propagation strength.

9. A personalized recommendation method as described in claim 8, characterized in that, The process of obtaining users' social relationship strength, social influence score, and interest similarity, and calculating the first interest propagation factor, includes: Obtain users' social influence scores, interest similarity, and social relationship strength; Based on a preset first balance parameter, the strength of the social relationship is logarithmically transformed to obtain the first propagation factor; The social influence score is subjected to a power transformation based on a preset second balance parameter to obtain the second propagation factor; Based on the preset third balance parameter, the interest similarity is exponentially transformed to obtain the third propagation factor; Based on the first propagation factor, the second propagation factor, and the third propagation factor, the first interest propagation factor is obtained.

10. A personalized recommendation method as described in claim 1, characterized in that, The step of calculating each user's interest score for a specific topic based on the interest propagation intensity and the interest topic distribution includes: Based on the distribution of interest topics and user behavior data, calculate the user's overall interest in the topics; Based on the user's social relationships, the user's community can be determined. The community interest influence is obtained based on the intensity of interest dissemination among each user in the community. Based on the overall interest level, the community interest influence, and the interest dissemination intensity, calculate each user's interest score for a specific topic.

11. A personalized recommendation method as described in claim 10, characterized in that, The step of calculating the user's overall interest in a topic based on the distribution of interest topics and user behavior data includes: Based on user behavior data, the number of user interaction behaviors in each topic is obtained; Based on the number of interaction behaviors, the topic attractiveness ratio of each topic to the user is obtained; Based on the topic attractiveness ratio and the user's interest topic distribution, calculate the user's overall interest in the topic.

12. The personalized recommendation method as described in claim 11, characterized in that, The process of determining the topic attractiveness ratio of each topic to the user based on the number of interaction behaviors includes: Obtain preset base weights and preset time decay factors for different behavior types; the behavior types include browsing, liking, commenting, and sharing; Calculate the average number of interactions per user across each behavior type based on the number of interactions described for each behavior type. Calculate the update weight for each behavior type based on the preset base weight, the preset time decay factor, and the average number of interactions. Based on the update weights and the number of user interactions under each behavior type, the topic attractiveness ratio of each topic to the user is calculated.

13. The personalized recommendation method as described in claim 10, characterized in that, The process of obtaining the community interest influence based on the interest dissemination intensity of each user in the community includes: The community interest centrality is obtained based on the interest propagation intensity of each user in the social network and the community. Based on the behavioral data of users in the community, the interest similarity between users, and the strength of social relationships, calculate the social influence score of each user in the community; The community interest influence is obtained based on the community interest centrality and the social influence score of each user in the community.

14. The personalized recommendation method as described in claim 10, characterized in that, The calculation of each user's interest score for a specific topic based on the overall interest level, the community interest influence, and the interest dissemination intensity includes: The community interest weight is obtained based on the dissemination of each user's interest and the influence of the community interest. Based on the distribution of user interest topics, user interest weights are obtained; Based on the community interest weight and the user interest weight, the user's overall interest score is weighted to obtain the user's interest score for a specific topic.

15. A personalized recommendation method as described in claim 1, characterized in that, The step of generating recommendation results for each user based on the interest score includes: Based on users' interest scores on different topics, target communities are selected from the current community set and used as recommendations for users.

16. A personalized recommendation system, characterized in that, include: The social relationship strength calculation module is used to acquire behavioral data of each user and calculate the strength of social relationships between users who have social relationships. The interest similarity calculation module is used to extract the distribution of interest topics of each user from the behavioral data and calculate the interest similarity between users; The interest propagation intensity calculation module is used to calculate the interest propagation intensity between users based on the interest similarity and the social relationship strength. The interest score calculation module is used to calculate the interest score of each user for a specific topic based on the interest propagation intensity and the interest topic distribution. The recommendation result generation module is used to generate recommendation results for each user based on the interest score.

17. A personalized recommendation device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the personalized recommendation method as described in any one of claims 1 to 15.

18. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the personalized recommendation method as described in any one of claims 1 to 15.

19. A computer program product, characterized in that, The computer program product includes a computer program or computer instructions, which, when executed by a processor, perform the personalized recommendation method as described in any one of claims 1 to 15.

Images