Cultural recommendation system and method based on user portrait

Abstract

The invention discloses a culture recommendation system and method based on a user portrait, and belongs to application of a data analysis and machine learning technology in the culture field, and the method comprises the following steps: collecting user data and culture content data, and constructing a comprehensive portrait of a user; the first recommendation system generates a preliminary recommendation result, optimizes weight parameters and optimizes a recommendation list; the second recommendation system generates secondary recommendation content by adopting user clustering and similarity analysis, and generates a final recommendation list by integrating the recommendation list of the first recommendation system; and displaying. The system comprises a data acquisition and processing module, an initial recommendation generation module, a user clustering and similarity analysis module and a recommendation result display module. According to the method, the basic information, the behavior characteristics and the emotional tendencies of the users are collected, the multi-dimensional user portraits are constructed, and the improved similarity calculation method and the recommendation algorithm are combined, so that the accuracy and the applicability of the recommendation system are improved, and the performance and the user satisfaction of the recommendation system are comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the application of data analysis and machine learning technologies in the cultural field, and specifically to a cultural recommendation system and method based on user portraits. Background Art

[0002] With the rapid development of information technology, the way of acquiring and disseminating cultural content has undergone tremendous changes, and the Internet has become the main channel. Faced with a massive amount of cultural content, how to effectively recommend high-quality content that meets the interests of users has become a hot topic and difficulty in current research. Traditional recommendation systems are mostly based on user behavior data and recommend content through similarity calculation. However, this recommendation method based on a single behavioral feature has some shortcomings, such as insufficient accuracy and diversity of recommendation results, and low user satisfaction.

[0003] As an emerging data analysis technology, user portrait technology can more accurately describe users' interests and needs by collecting and analyzing multi-dimensional data of users and constructing a comprehensive description of users' characteristics. The recommendation system combined with user portrait technology can understand users more comprehensively and provide more personalized and diversified recommendation services. However, the current recommendation system based on user portrait still has some problems and challenges in practical applications. For example, how to efficiently collect and process user data, how to build high-quality user portraits, and how to improve the accuracy and applicability of recommendation systems. Summary of the invention

[0004] The technical solution of the present invention aims at the technical problem that the existing technical solutions are too single, and provides a solution that is significantly different from the existing technologies. It mainly provides a cultural recommendation system and method based on user portraits, which is used to solve some problems and challenges existing in the actual application of the current recommendation system based on user portraits proposed in the above background technology.

[0005] The technical solution adopted by the present invention to solve the above technical problems is:

[0006] A cultural recommendation method based on user portraits includes the following steps:

[0007] First, collect user data and cultural content data to build a comprehensive user profile;

[0008] Then, the similarity of cultural content is calculated through the first recommendation system, and the user interest tags are matched to generate preliminary recommendation results, and then the weight parameters in the similarity calculation are optimized, and an optimized recommendation list is generated;

[0009] Finally, the recommendation list is presented to the customer.

[0010] As an improvement to the above solution, after the first recommendation system generates a recommendation result, the second recommendation system uses user clustering and similarity analysis to generate secondary recommendation content, which is then combined with the recommendation list of the first recommendation system to generate a final recommendation list; the recommendation list is then displayed to the customer.

[0011] Specifically, the user data includes the user's basic information, behavior data and emotional tendencies;

[0012] The basic information of the user includes the basic data of user registration information and usage log data; the behavior data includes the user's search, reading, clicking, browsing, collection and forwarding behavior data recorded by the backend database; the emotional tendency is the cultural field of user preference obtained by extracting the user's reading record from the log data;

[0013] And / or, the cultural content data includes cultural related information of articles and short videos.

[0014] Specifically, when constructing a comprehensive portrait of a user, the user's basic information, behavioral data, interest preferences, social relationships, and emotional tendencies are combined with multi-dimensional features to generate a comprehensive portrait of the user.

[0015] Specifically, the formula for calculating the similarity of cultural content is expressed as:

[0016]

[0017] Among them, Sim(a i ,a j ) is the similarity between cultural contents, ω k is the weight of the kth feature; f k (a i ,a j ) is the similarity of the kth feature between articles. The keyword weight vector of the article is extracted using TF-IDF, and the cosine similarity of the two articles on the keyword vector is calculated. The formula is expressed as:

[0018]

[0019] Among them A ai,k Indicates article a i TF-IDF weight on the kth word; Indicates article a j TF-IDF weight on the kth word;

[0020] And / or, the user interest matching formula is:

[0021]

[0022] where s uarepresents the predicted score of user u for content a, Sim(a,a′) represents the similarity between the contents, s ca′ represents the rating of article a′ by user u, and r(u) represents the collection of cultural articles and videos watched by user u.

[0023] Specifically, in the cultural content similarity formula, ω k The weight parameters are optimized using the particle swarm optimization algorithm. By simulating the flight trajectory of particles in the search space, the parameter combination that makes the objective function reach the optimal value is found. The calculation of the fitness value takes into account accuracy, recall rate and user satisfaction. The user satisfaction formula is expressed as:

[0024]

[0025] C represents customer experience, which can be quantified through user feedback, questionnaires, or satisfaction ratings. For example, user satisfaction can be scored on a scale of 1 to 5 through questionnaires. E represents expected value, which can be used to infer user expectations for response time, interface friendliness, etc. through industry standards or survey results of competing products. T represents technical capabilities, which can be measured through system performance indicators, such as system response time, stability, and throughput. ∑R represents the total number of recommended articles, N represents the number of adopted articles, and K represents knowledge coverage. Used to measure the performance of actual experience relative to user expectations. It reflects the degree to which technical capabilities support user experience. It reflects the overall acceptance rate of the recommended content. A higher value means that the recommended content is more in line with the user's interests and needs. It is used to measure the impact of knowledge coverage on user experience. The more comprehensive the knowledge coverage, the better the user experience. n1, n2, n3, and n4 represent weight parameters, which are used to balance the impact of different factors. Here, the A / B test method is used to continuously adjust the weights and observe the impact of different combinations on user satisfaction, and then find the appropriate weight settings.

[0026] The fitness function value formula is:

[0027] f(w)=α×ACC(w)+β×REC(w)+γ×U s

[0028] where w=(w1,w2…w n ), that is, the weight parameter to be optimized, ACC(w) represents accuracy, REC(w) represents recall, a,β,r∈{0.1,0.2…1.0}, select the parameter combination a,β,r that makes the objective function reach the optimal value;

[0029] And / or, the user interest matching formula with optimized parameters is:

[0030]

[0031] w is the optimized weight parameter.

[0032] Specifically, the specific steps of the second recommendation system using user clustering and similarity analysis are:

[0033] (1) First, use the clustering algorithm to simply classify users with similarity and count the number of predicted scores of cultural content for each user. i , the formula is:

[0034]

[0035] Where Count m Indicates the number of predicted scores corresponding to the i-th user, indicating whether a user has a predicted score for a specific cultural content. If the i-th user has a predicted score for the n-th cultural content, the value is 1, otherwise it is 0;

[0036] (2) Sort in descending order according to the number of predicted ratings, and select the top k users as the initial center points, denoted as {u1,u2,…,u k},and By selecting users with more rating data as the initial center point, and then executing the K-means algorithm for clustering, it can be expressed as:

[0037]

[0038] C j represents the jth cluster, represents the predicted rating vector of the i-th user; where u j It is the first The center point of the cluster, u l Indicates the center point of other clusters

[0039] (3) Recalculate the center point of each cluster,

[0040]

[0041] (4) Repeat the core iteration process of the clustering algorithm until the center point no longer changes and multiple clusters C are obtained. j gather;

[0042] (5) In each cluster, identify users who are similar to the target user. The formula is expressed as:

[0043]

[0044] Where P tk With P ik For user t and user i's predicted score for content k, and is the average rating of user t and user i, and α∈[0,1] is the weight parameter.

[0045] Furthermore, according to the similarity calculated by the first recommendation system, the top N users with the highest similarity to the target user are selected as the similar user set s N ; Based on similar user sets s N Based on the behaviors and preferences of target users, a recommendation list is generated for the target users. The content viewing records of similar users are summarized, the number of times each content is viewed is counted, and the recommendation score is calculated for each content. The formula is:

[0046]

[0047] where r u is the recommendation score, u i represents the i-th user, P ij represents the predicted score of user i for article j; S N Represents a set of similar users, usually the first few users most similar to the target user;

[0048] Sort the content according to the recommendation score, select the top M pieces of content with the highest scores, and generate secondary recommended content.

[0049] The present invention also provides a cultural recommendation system based on user portraits, which is applied to the above-mentioned cultural recommendation method based on user portraits. The recommendation system includes:

[0050] Data collection and processing module, used to collect user data and cultural content data, perform pre-processing and standardization, and build a comprehensive profile of users;

[0051] The initial recommendation generation module is used to calculate the similarity of cultural content, identify content that matches the user's interests, and match the user's interest tags to generate recommendations;

[0052] The user clustering and similarity analysis module is used to perform cluster analysis on users, identify users similar to the target user, and identify the content that the target user may be interested in through analysis of similar users; and generate the final recommendation list by combining the recommendation results of the initial recommendation generation module;

[0053] The recommendation result display module is used to display the recommendation list to the user.

[0054] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned cultural recommendation method based on user portrait when executing the computer program.

[0055] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned cultural recommendation method based on user portraits.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] (1) The first recommendation system of the present invention first generates preliminary recommendation results based on the similarity of cultural content and matches user interest tags. After the weight parameters of the similarity calculation are optimized, an optimized recommendation list is generated. In addition, the calculation of the fitness value in the value optimization not only considers accuracy and recall rate but also considers user satisfaction, thereby improving the accuracy and relevance of the recommendation results. The optimized recommendation system can better meet the personalized needs of users and significantly improve user satisfaction.

[0058] (2) In the optimization scheme of the present invention, a second recommendation system is provided based on the first recommendation system. The first recommendation system recommends similar content based on the user's own behavior, and the second recommendation system recommends based on the similarity between users. Then, the recommendation results of the two are combined for recommendation. Not only the personalized needs of users are considered, but also the similarity between users is combined, realizing the organic combination of personalized and group similarity recommendations, improving the diversity and matching degree of recommended content, and further improving the accuracy and applicability of the recommendation system.

[0059] (3) The present invention also provides specific operations for collecting user data and cultural content data and subsequent preprocessing and standardization processing. It uses text analysis technology to extract features such as themes, keywords, emotional tendencies, etc. in article or short video introductions, thereby achieving efficient collection and processing of relevant data.

[0060] (4) Before constructing a user portrait, the present invention converts the user's browsing history, articles read or video introductions, etc. into feature vectors through TF-IDF, and combines the user's basic information, behavioral characteristics, interest preferences, social relationships, emotional tendencies and other multi-dimensional features to generate a comprehensive user portrait, thereby achieving the construction of a high-quality user portrait. The user portrait combines basic information, behavioral characteristics and emotional tendencies, which can more comprehensively characterize the user's interests and generate more accurate recommendation results.

[0061] (5) The present invention introduces an improved similarity calculation method and recommendation algorithm, combined with user profiling technology and text analysis technology, to achieve more refined recommendations. The optimized algorithm not only improves the accuracy of the model, but also has better robustness and generalization capabilities. The system design of the present invention is simple, easy to implement and deploy, and has good scalability and maintainability. It can be flexibly adjusted according to actual needs to meet the needs of cultural personalized recommendations in different scenarios.

[0062] In summary, the present invention proposes a cultural recommendation system and method based on user portraits, which constructs a multi-dimensional user portrait by collecting users' basic information, behavioral characteristics, and emotional tendencies. Combined with an improved similarity calculation method and recommendation algorithm, the accuracy and applicability of the recommendation system are improved, thereby comprehensively improving the performance of the cultural recommendation system and user satisfaction.

[0063] The present invention will be explained in detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 An overall flow chart of a cultural recommendation method based on user portrait provided in the second embodiment of the present invention;

[0065] Figure 2 A flowchart of data collection and processing and user portrait formation in the present invention;

[0066] Figure 3 is a workflow diagram of the first recommendation system in the present invention;

[0067] Figure 4 This is a workflow diagram of the second recommendation system in the present invention. DETAILED DESCRIPTION

[0068] To facilitate understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings, but the present invention can be implemented in different forms and is not limited to the embodiments described in the text. On the contrary, these embodiments are provided to make the content disclosed in the present invention more thorough and comprehensive.

[0069] Example 1: Please refer to Figure 1-Figure 4 , a cultural recommendation method based on user portrait, including the following steps:

[0070] S1. First collect user and cultural content data, then process the user and cultural content data, and then build a user portrait.

[0071] Please refer to the attached Figure 2 , specifically:

[0072] (1) Multi-dimensional data collection;

[0073] Collect user data, including basic information, behavior data and emotional tendencies of users. Specifically, it includes: basic data of user registration information, use of log data and backend database to record users' search, reading, clicking, browsing, collection and forwarding behaviors, as well as extracting users' reading records through log data to obtain users' preferred cultural fields.

[0074] Collect cultural content data, including articles, short videos and other cultural related information. Specific sources include: online databases such as news websites, online magazines, video websites, etc., related websites and platforms such as Weibo culture-related public accounts, museum websites, cultural institution websites, etc.

[0075] (2) Data preprocessing;

[0076] Preprocess (mainly data cleaning) and standardize the collected user data and cultural content data. Use text analysis technology to extract themes, keywords, sentiment tendencies and other features in the article or short video introduction. You can use Python's NLTK or jieba library to decompose the article or related introduction content into single words or phrases, remove meaningless words in text analysis, use NLTK or spaCy library to mark the part of speech of each word, and then use the LDA (Latent Dirichlet Allocation) model of the Gensim library to extract topics.

[0077] (3) Use TF-IDF to extract keywords for users reading articles or watching short videos, and combine multi-dimensional features such as users’ basic information, behavioral data, interest preferences, social relationships, and emotional tendencies to form a comprehensive profile of users;

[0078] Before building a user profile, the user's browsing history, read articles or video introductions, etc. are converted into feature vectors through TF-IDF. The formula is:

[0079]

[0080] Where count(t,d) represents the number of times word t appears in document d, count(d) represents the number of times all words appear in document d, N represents the total number of documents in the corpus, and N t Represents the number of documents containing the term t, which can be calculated using the sklearn library in Python.

[0081] By using methods such as TF-IDF to extract features from text data, user behavior data and interest preference data are converted into numerical features, and the user's interest model is constructed based on the user's behavior data and interest preference data. The functions of the interest model include:

[0082] ①User interest model is an important part of user portrait;

[0083] ② In the recommendation system, the user interest model is used to match user interest tags and generate a preliminary recommendation list;

[0084] ③ During the optimization process of the recommendation system, the system dynamically updates the user interest model through the user's latest behavior data, so that the user portrait can be continuously adjusted as the user's interests change, ensuring the accuracy and relevance of the recommended content.

[0085] Combine multi-dimensional features such as the user's basic information, behavioral characteristics, interest preferences, social relationships, and emotional tendencies to generate a comprehensive portrait of the user.

[0086] Through the above steps and methods, a comprehensive user profile can be constructed to describe the user's interests, behaviors, and preferences. This profile is used in a personalized recommendation system to improve the accuracy of recommendations and user satisfaction.

[0087] S2. The first recommendation system generates a preliminary recommendation list, optimizes the weight parameters in the similarity calculation, and generates an optimized recommendation list.

[0088] Please refer to the attached Figure 3 , specifically:

[0089] The first recommendation system of the present invention calculates the similarity of cultural content and matches user interest tags to generate preliminary recommendation results.

[0090] The recommendation algorithm is constructed based on culturally relevant content (articles, short videos, etc.). The text analysis technology is used to extract themes, keywords, emotional tendencies and other features in the article or short video introduction, and the similarity between cultural contents is calculated. The formula is expressed as:

[0091]

[0092] Among them, Sim(a i ,a j ) is the similarity between cultural contents, ω k is the weight of the kth feature; f k (a i ,a j ) is the similarity of the kth feature between articles. The keyword weight vector of the article is extracted using TF-IDF, and the cosine similarity of the two articles on the keyword vector is calculated. The formula is expressed as:

[0093]

[0094] Among them A ai,k Indicates article ai TF-IDF weight on the kth word; Indicates article a j TF-IDF weight on the k-th word.

[0095] According to the user's viewing history and interest tags, match cultural content that matches the user's interests, calculate the user's predicted score for each content, and match content that matches the user's interests. The user interest matching formula is:

[0096]

[0097] where s ua represents the predicted score of user u for content a, Sim(a,a′) represents the similarity between the contents, s ca′ represents the rating of article a′ by user u, and r(u) represents the collection of cultural articles and videos watched by user u.

[0098] After generating the preliminary recommendation list, replace ω in the above similarity formula k The weight parameters are optimized using the particle swarm optimization algorithm. By simulating the flight trajectory of particles in the search space, the parameter combination that makes the objective function reach the optimal value is found. The calculation of the fitness value not only considers accuracy and recall rate but also user satisfaction. The user satisfaction formula can be expressed as:

[0099]

[0100] C represents customer experience, which can be quantified through user feedback, questionnaires, or satisfaction ratings. For example, user satisfaction can be scored on a scale of 1 to 5 through questionnaires. E represents expected value, which can be used to infer user expectations for response time, interface friendliness, etc. through industry standards or survey results of competing products. T represents technical capabilities, which can be measured through system performance indicators, such as system response time, stability, and throughput. ∑R represents the total number of recommended articles, N represents the number of adopted articles, and K represents knowledge coverage. Used to measure the performance of actual experience relative to user expectations. It reflects the degree to which technical capabilities support user experience. It reflects the overall acceptance rate of the recommended content. A higher value means that the recommended content is more in line with the user's interests and needs. It is used to measure the impact of knowledge coverage on user experience. The more comprehensive the knowledge coverage, the better the user experience. n1, n2, n3, and n4 represent weight parameters, which are used to balance the impact of different factors. Here, the A / B test method is used to continuously adjust the weights and observe the impact of different combinations on user satisfaction, and then find the appropriate weight settings.

[0101] The fitness function value formula is:

[0102] f(w)=α×ACC(w)+β×REC(w)+γ×U s

[0103] where w=(w1,w2…w n ), that is, the weight parameter to be optimized, ACC(w) represents accuracy, REC(w) represents recall, a, β, r∈{0.1, 0.2…1.0}, and the parameter combination a, β, r that makes the objective function reach the optimal value is selected. If the fitness value of the current particle is better than the best fitness value previously encountered by the particle, the individual optimal position is updated, and the particle with the best fitness value is selected from all particles, and its position is used as the global optimal position. The speed and position update process is repeated until the predetermined number of iterations is reached or the fitness value converges, and the optimal weight parameter combination is obtained.

[0104] The user interest matching formula with optimized parameters is:

[0105]

[0106] w is the optimized weight parameter, which sorts the recommendation results and displays the most relevant content to users.

[0107] The first recommendation system of the present invention can calculate the similarity of cultural content based on user interest and behavior data using text analysis technology and generate a personalized recommendation list, thereby improving the accuracy of recommendations and user satisfaction.

[0108] S3. Recommendation result display: Display the most relevant cultural content to users.

[0109] The user portrait-based cultural recommendation system corresponding to the above method mainly includes a data collection and processing module, an initial recommendation generation module, a user clustering and similarity analysis module, and a recommendation result display module.

[0110] The data collection and processing module collects user behavior data, user registration information, usage log data and data recorded in the backend database. This module also collects cultural content and performs standardized processing. User portraits are formed based on user behavior, interests, cultural content and other characteristics.

[0111] The initial recommendation generation module constructs a recommendation algorithm based on user interests and behavior. In the process of constructing the recommendation algorithm, the initial recommendation generation module adopts methods such as TF-IDF (term frequency-inverse document frequency) and LDA (latent Dirichlet allocation). The initial recommendation generation module calculates the similarity between cultural contents. By calculating the similarity of cultural contents, the initial recommendation generation module can identify the content that matches the user's interests and match the user's interest tags to generate recommendations. Through this module, the system can achieve personalized and accurate content recommendations to meet the diverse needs of users.

[0112] The recommendation result display module displays the most relevant cultural content to the user.

[0113] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0114] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0115] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0116] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0117] Embodiment 2: This embodiment differs from Embodiment 1 in that:

[0118] In this embodiment, after the first recommendation system generates the recommendation results, the second recommendation system also uses user clustering and similarity analysis to generate secondary recommendation content, which is combined with the recommendation list of the first recommendation system to generate a final recommendation list for display to customers.

[0119] Please refer to the attached Figure 4 , specifically:

[0120] The prediction score matrix S is constructed based on the prediction score generated by the first recommendation system u , and the predicted score is standardized to obtain the standardized predicted score matrix S′ u , the formula is:

[0121]

[0122] Where m is the number of users, n is the number of cultural articles, represents the predicted rating of the mth user for the nth cultural content, μ n represents the average predicted score of the nth cultural content, σ nRepresents the standard deviation of the predicted score of the nth cultural content. Through the predicted score matrix, the similarity between users can be calculated, and the recommendation list can be further optimized based on this similarity. After calculating the similarity between users, users can be clustered and users with similar interests can be clustered in the same cluster. Within each cluster, the user who is most similar to the target user can be further identified. The behaviors and preferences of similar users are used to generate recommended content for the target user.

[0123] Use clustering algorithms to simply classify users with similarity and count the number of predicted cultural content scores for each user. i , the formula is:

[0124]

[0125] Where Count i It represents the number of predicted scores corresponding to the i-th user, flag∈(0,1) indicates whether a user has a predicted score for a specific cultural content. If the i-th user has a predicted score for the n-th cultural content, the value is 1, otherwise it is 0.

[0126] Sort in descending order according to the number of predicted ratings, and select the first k users as the initial center points, denoted as {u1,u2,…,u k},and By selecting users with more rating data as the initial center points, the rating data of these users is more comprehensive and accurate, and can better represent the preferences of the user group. The K-means algorithm is executed for clustering, which is expressed as:

[0127]

[0128] C j represents the jth cluster, represents the predicted rating vector of the i-th user. j It is the first The center point of the cluster, u l represents the center point of other clusters, u j with u l The center points of different clusters are used to determine which cluster each user should be assigned to.

[0129] Recalculate the center point of each cluster,

[0130]

[0131] Repeat the core iteration process of the K-means algorithm (first calculate the distance from the user to the cluster center, that is, the predicted rating vector S′ of each user u, calculate the distance between the user and the cluster center; then reallocate the user to the nearest cluster center) until the center point no longer changes. At this point, multiple users are roughly classified to obtain multiple clusters C j In each cluster, we further identify users who are similar to the target user. The formula is:

[0132]

[0133] Where P tk With P ik is the predicted score of user t and user i for content k (i.e., the predicted score s of the first recommendation system ua ), and is the average rating of user t and user i, and α∈[0,1] is the weight parameter.

[0134] As a preferred solution, according to the similarity calculated by the first recommendation system, the top N users with the highest similarity to the target user are selected as the similar user set s N Based on similar user sets s N Based on the behaviors and preferences of target users, a recommendation list is generated for the target users. The content viewing records of similar users are summarized, the number of times each content is viewed is counted, and the recommendation score is calculated for each content. The formula is:

[0135]

[0136] where r u is the recommendation score, u i represents the i-th user, P ij represents the predicted score of user i for article j; S N Represents a set of similar users, usually the first few users most similar to the target user. Sort the content according to the recommendation score, select the top M pieces of content with the highest score, combine them with the recommendation list of the first recommendation system, and sort them in descending order to generate the final recommendation list.

[0137] Correspondingly, the user portrait-based cultural recommendation system corresponding to the above method also includes a user clustering and similarity analysis module.

[0138] The user clustering and similarity analysis module performs cluster analysis on users, groups users with similar interests and behaviors into the same cluster, and further identifies users similar to the target user in each cluster. The module can identify users with the most similar interests and behaviors to the target user. The behaviors and preferences of these similar users can provide important references for generating recommendation lists. Based on the browsing records, click records, search records, etc. of similar users, the module can identify content that the target user may be interested in. In this process, user portraits play a key role. They integrate the user's behavioral data and interest tags to fully reflect the user's interest preferences. Through this module, the system can achieve personalized and precise content recommendations. It can also combine the recommendation results of the first recommendation system and the second recommendation system to obtain the final recommendation results, and sort the final recommendation results.

[0139] The rest is the same as the first embodiment.

[0140] The above is an exemplary description of the present invention in combination with the accompanying drawings. It is obvious that the specific implementation of the present invention is not limited to the above-mentioned method. As long as such non-substantial improvements are made by adopting the method concept and technical solution of the present invention, or the concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A cultural recommendation method based on user portrait, characterized by: The steps include: First, collect user data and cultural content data to build a comprehensive user profile; Then, the similarity of cultural content is calculated through the first recommendation system, and the user interest tags are matched to generate preliminary recommendation results, and then the weight parameters in the similarity calculation are optimized, and an optimized recommendation list is generated; Finally, the recommendation list is presented to the customer.

2. The cultural recommendation method based on user portrait according to claim 1, characterized in that: After the first recommendation system generates the recommendation results, the second recommendation system uses user clustering and similarity analysis to generate secondary recommendation content, which is then combined with the recommendation list of the first recommendation system to generate the final recommendation list; Then show the recommended list to the customer.

3. The cultural recommendation method based on user portrait according to claim 1, characterized in that: The user data includes the user's basic information, behavior data and emotional tendencies; The basic information of the user includes the basic data of user registration information and usage log data; the behavior data includes the user's search, reading, clicking, browsing, collection and forwarding behavior data recorded by the backend database; The emotional tendency is the cultural field of user preference obtained by extracting the user's reading records from log data; And / or, the cultural content data includes cultural related information of articles and short videos.

4. The method for cultural recommendation based on user portrait according to claim 3, characterized in that: When constructing a comprehensive portrait of a user, the user's basic information, behavioral data, interest preferences, social relationships, and emotional tendencies are combined in multiple dimensions to generate a comprehensive portrait of the user.

5. The cultural recommendation method based on user portrait according to claim 1, characterized in that: The formula for calculating the similarity of cultural content is expressed as: Among them, Sim(a i ,a j ) is the similarity between cultural contents, ω k is the weight of the kth feature; f k (a i ,a j ) is the similarity of the kth feature between articles. The keyword weight vector of the article is extracted using TF-IDF, and the cosine similarity of the two articles on the keyword vector is calculated. The formula is expressed as: Among them A ai,k Indicates article a i TF-IDF weight on the kth word; Indicates article a j TF-IDF weight on the kth word; And / or, the user interest matching formula is: where s ua represents the predicted score of user u for content a, Sim(a,a′) represents the similarity between the contents, s ca′ represents the rating of article a′ by user u, and r(u) represents the collection of cultural articles and videos watched by user u.

6. The method for cultural recommendation based on user portrait according to claim 5, characterized in that: In the formula of cultural content similarity, ω k The weight parameters are optimized using the particle swarm optimization algorithm. By simulating the flight trajectory of particles in the search space, the parameter combination that makes the objective function reach the optimal value is found. The calculation of the fitness value takes into account accuracy, recall rate and user satisfaction. The user satisfaction formula is expressed as: Among them, C represents customer experience, which is quantified through user feedback, questionnaires or satisfaction ratings; E represents expected value, which is inferred from user expectations through industry standards or survey results of competing products; T represents technical capabilities, which is measured through system performance indicators; ΣR represents the total number of recommended articles; N represents the number of adopted articles; K represents knowledge coverage; Used to measure the performance of actual experience relative to user expectations; Reflects the degree to which technical capabilities support user experience; Reflects the overall acceptance rate of recommended content. A higher value means that the recommended content is more in line with the user's interests and needs. It is used to measure the impact of knowledge coverage on user experience. The more comprehensive the knowledge coverage, the better the user experience. n1, n2, n3, and n4 represent weight parameters, which are used to balance the impact of different factors. The fitness function value formula is: f(w)=α×ACC(w)+β×REC(w)+γ×U s where w=(w1,w2…w n ), that is, the weight parameter to be optimized, ACC(w) represents accuracy, REC(w) represents recall, a,β,r∈{0.1,0.2…1.0}, select the parameter combination a,β,r that makes the objective function reach the optimal value; And / or, the user interest matching formula with optimized parameters is: w is the optimized weight parameter.

7. The method for cultural recommendation based on user portrait according to claim 2, characterized in that: The specific steps of the second recommendation system using user clustering and similarity analysis are as follows: (1) First, use the clustering algorithm to simply classify users with similarity and count the number of predicted scores of cultural content for each user. i , the formula is: Where Count m Indicates the number of predicted scores corresponding to the i-th user, flag∈(0,1) indicates whether a user has a predicted score for a specific cultural content. If the i-th user has a predicted score for the n-th cultural content, the value is 1, otherwise it is 0; (2) Sort in descending order according to the number of predicted ratings, and select the top k users as the initial center points, denoted as {u1,u2,…,u k },and By selecting users with more rating data as the initial center point, and then clustering, it can be expressed as: C j represents the jth cluster, represents the predicted rating vector of the i-th user; where u j It is the first The center point of the cluster, u l Indicates the center point of other clusters; (3) Recalculate the center point of each cluster, (4) Repeat the core iteration process of the clustering algorithm until the center point no longer changes and multiple clusters C are obtained. j gather; (5) In each cluster, identify users who are similar to the target user. The formula is expressed as: Where P tk With P ik For user t and user i's predicted score for content k, and is the average rating of user t and user i, and α∈[0,1] is the weight parameter.

8. The method for cultural recommendation based on user portrait according to claim 7, characterized in that: According to the similarity calculated by the first recommendation system, the top N users with the highest similarity to the target user are selected as the similar user set s N ; Based on similar user sets s N Based on the behaviors and preferences of target users, a recommendation list is generated for the target users, the content viewing records of similar users are summarized, the number of times each content is viewed is counted, and the recommendation score is calculated for each content. The formula is: where r u is the recommendation score, u i represents the i-th user, P ij represents the predicted score of user i for article j; S N Represents a set of similar users; Then sort the content according to the recommendation score, select the top M pieces of content with the highest scores, and generate secondary recommended content.

9. A cultural recommendation system based on user portraits, characterized by: The cultural recommendation method based on user portrait according to any one of claims 1 to 8, wherein the recommendation system comprises: Data collection and processing module, used to collect user data and cultural content data, perform pre-processing and standardization, and build a comprehensive profile of the user; The initial recommendation generation module is used to calculate the similarity of cultural content, identify content that matches the user's interests, and match the user's interest tags to generate recommendations; The user clustering and similarity analysis module is used to perform cluster analysis on users, identify users similar to the target user, and identify the content that the target user may be interested in through analysis of similar users; and generate the final recommendation list by combining the recommendation results of the initial recommendation generation module; The recommendation result display module is used to display the recommendation list to the user.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the cultural recommendation method based on user portrait described in any one of claims 1-8 are implemented.

Images