Recommendation system mixed filling optimization method and system fusing user information and historical score data
By extracting and clustering users' feature and regularly filling them with historical scoring data, the data sparsity and cold start problems in the recommendation system are solved, and the accuracy and personalization of recommendations are improved.
Patent Information
- Application Number
- CN202411673303.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-21
- Publication Date
- 2025-02-21
AI Technical Summary
When facing data sparseness and cold start problems, existing recommendation systems are difficult to effectively utilize user characteristic information, resulting in insufficient recommendation accuracy and personalization.
By extracting and encoding the user's personalized information, the user is divided into clusters with similar characteristics using K-means clustering, and combining historical scoring data, a regularization strategy is used to balance the average cluster score and the average user's historical scoring score. to fill in missing values in the scoring matrix.
It significantly improves the filling accuracy and personalization of the recommendation system, effectively alleviates data sparsity and cold start problems, and enhances the adaptability and user experience of the recommendation system.
Smart Images
Figure CN119493910A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to but is not limited to the technical field of recommendation systems, and in particular relates to a hybrid filling optimization method and system that integrates user information and historical rating data. Background Art
[0002] With the rapid development of big data, e-commerce and Internet applications, recommendation systems have been widely used in the fields of movies, music, commodities, etc. Faced with the diversification of user needs and the continuous growth of data scale, recommendation systems have also exposed a series of technical challenges. Traditional collaborative filtering algorithms, such as user-based or item-based collaborative filtering methods, usually rely on user-item rating matrices. However, with the rapid increase in the number of users and items, the sparsity problem of the rating matrix has become more serious, resulting in a significant decrease in the accuracy of the recommendation system.
[0003] Some studies have tried to alleviate the impact of data sparsity. For example, the SVD (singular value decomposition) method proposed by Sarwar et al. reduces the dimensionality of the rating matrix to reduce data sparsity, thereby improving the accuracy of the recommendation system. However, the SVD method has a high computational complexity and is slow to process large-scale data sets. In addition, the SVD method mainly relies on existing rating data and does not effectively utilize other feature information of users, so the improvement of the cold start problem is limited. Another commonly used filling method is to use mean filling, that is, to fill the unrated items with the average rating of the item or user. Ahmadian et al. added rating correction for unreliable users on the basis of mean filling, thereby further optimizing the accuracy of the predicted rating. However, the mean filling method usually cannot reflect the preference differences of individual users, and it is easy to cause rating distortion when the rating behaviors of users vary greatly.
[0004] In view of the above analysis, the technical problems that urgently need to be solved in the existing technology are: the algorithms proposed in the existing technology for data sparsity have solved the sparsity problem of the rating matrix to a certain extent, but they need to rely on a large amount of external data or complex model training, ignore the diversity of user characteristics, and there may be errors in rating prediction. Summary of the invention
[0005] In view of the problems existing in the prior art, the present invention provides a hybrid filling optimization method and system that integrates user information and historical rating data.
[0006] The present invention is implemented as follows: a hybrid filling optimization method for a recommendation system integrating user information and historical rating data, comprising the following steps:
[0007] S1, extract and encode the user's personalized information, and then use K-means to divide the users into several clusters with similar features.
[0008] S2, performs statistical analysis on the rating behavior in each cluster, calculates the average rating of items in the cluster, and also calculates the historical average rating of each user.
[0009] S3, through the regularization strategy, balances the user's historical average rating with the cluster average rating, fills in the missing values in the rating matrix, and generates a more complete rating matrix. The similarity is calculated using the filled rating matrix, and the K users with the highest similarity are selected as neighbors based on the calculation results. The ratings of the target items by the neighbor users are used to predict the target user's ratings for the unrated items, and the N items with the highest ratings are selected as recommendations for the target user.
[0010] Further, the S1 specifically includes:
[0011] In the first step, the user age is divided into four intervals: 1-18 years old (teenagers), 19-35 years old (youth), 36-50 years old (middle-aged), and over 50 years old (elderly). Each interval is coded with the numbers 0, 1, 2, and 3 to form part of the age feature vector.
[0012] In the second step, the gender feature is binarized, with male represented by 1 and female represented by 0, as an element in the feature vector.
[0013] The third step is to divide occupations into four categories: professional occupations (such as doctors and lawyers), cultural occupations (such as teachers and artists), social occupations (such as civil servants and salespeople), and other occupations. Each type of occupation is coded with the numbers 0, 1, 2, and 3, forming part of the occupational feature vector.
[0014] The fourth step is to encode the features of each user into a three-dimensional vector and use the K-means algorithm to perform clustering.
[0015] Further, the S2 specifically includes:
[0016] The first step is to find similar user clusters based on user characteristics when filling in the ratings for the target users.
[0017] The second step is to count the scores of items in each cluster and calculate the average score R of the items in the cluster. c At the same time, a rating ratio threshold is set to ensure that only items with popularity exceeding the threshold can be used for cluster average rating calculation. I and user set S U The ratio of the number of users who have rated a project to the total number of users in the project score data is the project popularity of the project, which is expressed as:
[0018]
[0019] Among them, if user u scores item i, then λ u,i =1, otherwise 0. |S U | represents the total number of people in the project rating data, H i Indicates the popularity of the project, and its value range is between [0,1]. i The larger the value, the more popular the item.
[0020] The third step is to take the average rating of each user on the rated items as the user's historical average rating R u .
[0021] Further, the S3 specifically includes:
[0022] In the first step, after calculating the cluster rating and user history rating, the regularization method is used to combine the two to fill the missing values in the rating matrix. It is expressed as:
[0023]
[0024] Where: P u,m R represents the filling score (predicted score) of user u on item m. u represents the average rating of user u, R c It represents the average rating of item m by other users in the cluster, and it represents the collective rating preference of users in the cluster. c represents the rating deviation of cluster users on item m. λ is the regularization coefficient, which is used to balance the user's own rating preference and the collective rating within the cluster.
[0025] The second step is to use the filled rating matrix to calculate the similarity. Based on the similarity calculation results, select the K users with the highest similarity as neighbors.
[0026] The third step is to use the ratings of the target items by neighbor users to predict the ratings of the target users on the unrated items, which is expressed as:
[0027]
[0028] in The predicted score of user u for item i, is the average rating of users, r v,i The actual rating given to item i by similar user v, The average rating of similar users v, N u is a set of users similar to user u, and v∈N u is one of these similar users. sim(u,v) is the similarity between user u and user v.
[0029] The fourth step is to select the N items with the highest scores according to the predicted scores as recommendations for the target users.
[0030] Another object of the present invention is to provide a hybrid filling optimization system for integrating user information and historical rating data to implement the hybrid filling optimization method integrating user information and historical rating data, comprising:
[0031] The user clustering module extracts and encodes the user's personalized information, and then uses K-means to divide the users into several clusters with similar features.
[0032] The historical average rating calculation module performs statistical analysis on the rating behavior in each cluster, calculates the average rating of items in the cluster, and also calculates the historical average rating of each user.
[0033] The target user recommendation module uses a regularization strategy to balance the user's historical average rating with the cluster average rating, fill in the missing values in the rating matrix, and generate a more complete rating matrix. The similarity is calculated using the filled rating matrix, and the K users with the highest similarity are selected as neighbors based on the calculation results. The ratings of the target items by the neighbor users are used to predict the target user's ratings for the unrated items, and the N items with the highest ratings are selected as recommendations for the target user.
[0034] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the hybrid filling optimization method for the recommendation system that integrates user information and historical rating data.
[0035] Another object of the present invention is to provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the hybrid filling optimization method of the recommendation system that integrates user information and historical rating data.
[0036] Another object of the present invention is to provide an information data processing terminal, which includes the recommendation system hybrid filling optimization system that integrates user information and historical rating data.
[0037] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:
[0038] First, in the recommendation system, the sparsity of the rating matrix is an urgent problem to be solved. With the increase in the number of users and items, most users only rate a small number of items, resulting in a large number of null values in the matrix, which affects the system performance. Sparsity not only makes the similarity calculation inaccurate, but also aggravates the cold start problem and limits the recommendation effect of new users and new items. Traditional filling algorithms are insufficient in dealing with data sparsity and personalized needs. The present invention proposes an innovative filling method based on user personalized features and historical scores, which effectively alleviates the sparsity problem of the rating matrix through K-means clustering and regularization strategies. First, the K-means clustering algorithm is used to cluster users according to features such as age, gender, and occupation to make full use of the similarities between users. The cluster score provides a group preference reference for filling and improves the accuracy of similarity calculation. Secondly, through the regularization strategy, the cluster score is balanced with the user's historical score, so as to comprehensively consider individual preferences and group commonalities and accurately fill the missing values of the matrix. Compared with the global mean or simple interpolation filling, this method can better characterize individual differences among users and further improve the filling stability. In practical applications, with the help of cluster information and regularization processing, the present invention significantly improves the filling accuracy, effectively alleviates the cold start problem, enhances the recommendation effect and adaptability of the system, and makes the recommendation more personalized and accurate.
[0039] Second, as auxiliary evidence of the inventiveness of the claims of the present invention, it is also reflected in the fact that the technical solution of the present invention solves the problems of data sparsity and cold start in the recommendation system to a certain extent:
[0040] Traditional collaborative filtering recommendation algorithms perform poorly in data sparsity and cold start problems, making it difficult to accurately capture the personalized needs of users, resulting in a narrow coverage of recommended items, making it difficult to meet the diverse preferences of users, and affecting the actual recommendation effect. Unlike other data filling algorithms, the present invention combines user feature information with historical rating data, and optimizes the recommendation process through feature clustering and regularization filling methods. Specifically, the personalized information of users (such as age, gender, occupation, etc.) is first extracted, and K-means clustering is performed on users based on these features to form user clusters with similar features. Subsequently, the average score of each item is calculated within the cluster, and combined with the user's historical rating, a regularization strategy is used to balance the cluster average score with the individual score to complete the missing values in the rating matrix. This method of integrating user feature information and historical ratings enables the recommendation system to capture user preferences more accurately, especially in data sparseness and cold start scenarios, and can still generate effective recommendations. By balancing individual preferences and group characteristics in the recommendation process, the accuracy of the recommendation is improved, while the diversity of the recommended items is increased, so that the recommendation results are closer to the actual needs of the user, thereby enhancing the effect and applicability of the recommendation system in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 This is a step diagram of a hybrid filling optimization method for a recommendation system that integrates user information and historical rating data provided by an embodiment of the present invention;
[0042] Figure 2 This is a flow chart of a hybrid filling optimization method for a recommendation system that integrates user information and historical rating data provided by an embodiment of the present invention;
[0043] Figure 3 It is a structural diagram of a hybrid filling optimization system of a recommendation system that integrates user information and historical rating data provided by an embodiment of the present invention.
[0044] Figure 4-5 It is relevant evidence of the technical effects obtained by the embodiments of the present invention. DETAILED DESCRIPTION
[0045] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0046] See also Figure 1 As shown, the method step diagram of the embodiment of the present invention includes the following steps:
[0047] S101: Extract and encode the user's personalized information, and then use K-means to divide the users into several clusters with similar features.
[0048] S102: Statistically analyze the rating behaviors in each cluster, calculate the average rating of the items in the cluster, and also calculate the historical average rating of each user.
[0049] S103: Through the regularization strategy, the user's historical average rating is balanced with the cluster average rating, the missing values in the rating matrix are filled, and a more complete rating matrix is generated. The similarity is calculated using the filled rating matrix, and the K users with the highest similarity are selected as neighbors based on the calculation results. The ratings of the target items by the neighbor users are used to predict the ratings of the target user for the unrated items, and the N items with the highest ratings are selected as recommendations for the target user.
[0050] The method of the present invention realizes personalized recommendation through a series of steps, mainly including feature extraction, cluster analysis, score filling and recommendation prediction. The specific principles are as follows:
[0051] First, in step S101, the system extracts and encodes the user's personalized information (such as age, interests, usage habits, etc.) to convert the user's different features into a computable vector form. By using the K-means clustering algorithm, users are divided into several clusters based on the similarity of their features, so that users with similar preferences and features are grouped together for further analysis. This processing can not only reduce the computational complexity, but also improve the accuracy of recommendations.
[0052] Next, in step S102, the system performs statistical analysis on the rating behaviors in each cluster, calculates the average rating of items in each cluster, and records the historical average rating of each user. Through statistical analysis within the cluster, the preference patterns and rating tendencies of different user groups can be identified, providing basic data for subsequent rating matrix filling and recommendation prediction.
[0053] In step S103, in order to overcome the problem of missing data in the rating matrix, the system adopts a regularization strategy to balance the user's historical average rating and the cluster average rating, thereby filling the missing values in the rating matrix. This balancing strategy can more accurately infer the user's potential preference for unrated items, generate a more complete rating matrix, and make subsequent similarity calculations and recommendation predictions more accurate.
[0054] After that, the similarity between users is calculated using the filled rating matrix. By analyzing the similarity of different users' rating behaviors, the system selects the K users with the highest similarity to the target user as neighbor users. This similarity analysis is based on the rating matrix and can help the system identify neighbors with the closest interests to the target user, providing a reliable data basis for recommendation prediction.
[0055] In the recommendation prediction stage, the system predicts the target user's possible ratings for unrated items based on the ratings of the target items by neighbor users, and sorts them according to the ratings. Finally, the system selects the N items with the highest ratings and presents them to the target user as the recommendation results. This prediction based on neighbor ratings can effectively improve the relevance of the recommended content.
[0056] Through the above steps, the method of the present invention can generate personalized recommendations, which not only improves the accuracy of the recommendations, but also enhances the user experience, allowing users to more conveniently find items that meet their personal preferences.
[0057] The following is combined with Figure 2 The application principle of the present invention is further described.
[0058] Figure 2 For the specific implementation process, the present invention provides a hybrid filling optimization algorithm that integrates user information and historical rating data, including the following steps:
[0059] 1. Perform feature extraction and encoding on the user's personalized information, and then use K-means to divide the users into several clusters with similar features.
[0060] 1.1 The user age is divided into four intervals: 1-18 years old (teenagers), 19-35 years old (youth), 36-50 years old (middle-aged), and over 50 years old (elderly). Each interval is coded with the numbers 0, 1, 2, and 3, forming part of the age feature vector.
[0061] 1.2 Binarize the gender feature, with male represented by 1 and female represented by 0, as an element in the feature vector.
[0062] 1.3 Occupations are divided into four categories: professional occupations (such as doctors and lawyers), cultural occupations (such as teachers and artists), social occupations (such as civil servants and salespeople) and other occupations. Each type of occupation is coded with the numbers 0, 1, 2, and 3, forming part of the occupational feature vector.
[0063] 1.4 Encode the features of each user into a three-dimensional vector and use the K-means algorithm for clustering.
[0064] 2 Perform statistical analysis on the rating behavior in each cluster, calculate the average rating of items in the cluster, and also calculate the historical average rating of each user.
[0065] 2.1 When filling in the ratings for target users, first find similar user clusters based on user characteristics.
[0066] 2.2 Count the scores of items in each cluster and calculate the average score R of the items in the cluster c At the same time, a rating ratio threshold is set to ensure that only items with popularity exceeding the threshold can be used for cluster average rating calculation. I and user set S U The ratio of the number of users who have rated a project to the total number of users in the project rating data is the project popularity of the project, which is expressed as:
[0067]
[0068] Among them, if user u scores item i, then λ u,i =1, otherwise 0. |S U | represents the total number of people in the project rating data, H i Indicates the popularity of the project, and its value range is between [0,1]. i The larger the value, the more popular the item.
[0069] 2.3 Take the average rating of each user on the rated items as the user's historical average rating R u .
[0070] 3. Through the regularization strategy, the user's historical average rating is balanced with the cluster average rating, the missing values in the rating matrix are filled, and a more complete rating matrix is generated. The similarity is calculated using the filled rating matrix, and the K users with the highest similarity are selected as neighbors based on the calculation results. The ratings of the target items by the neighbor users are used to predict the target user's ratings for the unrated items, and the N items with the highest ratings are selected as recommendations for the target user.
[0071] 3.1 After calculating the cluster rating and user history rating, the regularization method is used to combine the two to fill the missing values in the rating matrix. It is expressed as:
[0072]
[0073] Where: P u,m R represents the filling score (predicted score) of user u on item m. u represents the average rating of user u, R c It represents the average rating of item m by other users in the cluster, and it represents the collective rating preference of users in the cluster. c represents the rating deviation of cluster users on item m. λ is the regularization coefficient, which is used to balance the user's own rating preference and the collective rating within the cluster.
[0074] 3.2 Use the filled rating matrix to calculate the similarity. According to the similarity calculation results, select the K users with the highest similarity as neighbors.
[0075] 3.3 Using the ratings of neighbor users on the target item, predict the target user's ratings on the unrated item. It is expressed as:
[0076]
[0077] in The predicted score of user u for item i, is the average rating of users, r v,i The actual rating given to item i by similar user v, The average rating of similar users v, N u is a set of users similar to user u, and v∈N u is one of these similar users. sim(u,v) is the similarity between user u and user v.
[0078] 3.4 Based on the predicted scores, select the N items with the highest scores as recommendations for the target users.
[0079] The above are only preferred examples of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
[0080] like Figure 3 As shown, the hybrid filling optimization system of the recommendation system integrating user information and historical rating data provided by the embodiment of the present invention includes:
[0081] The user clustering module extracts and encodes the user's personalized information, and then uses K-means to divide the users into several clusters with similar features.
[0082] The historical average rating calculation module performs statistical analysis on the rating behavior in each cluster, calculates the average rating of items in the cluster, and also calculates the historical average rating of each user.
[0083] The target user recommendation module uses a regularization strategy to balance the user's historical average rating with the cluster average rating, fill in the missing values in the rating matrix, and generate a more complete rating matrix. The similarity is calculated using the filled rating matrix, and the K users with the highest similarity are selected as neighbors based on the calculation results. The ratings of the target items by the neighbor users are used to predict the target user's ratings for the unrated items, and the N items with the highest ratings are selected as recommendations for the target user.
[0084] An application embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of a hybrid filling optimization method that integrates user information and historical rating data.
[0085] An application embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of a hybrid filling optimization method that integrates user information and historical rating data.
[0086] An application embodiment of the present invention provides an information data processing terminal, which includes a hybrid filling optimization system that integrates user information and historical rating data.
[0087] The method of the present invention has a wide range of application values in multiple fields. In e-commerce platforms (such as Taobao), the present invention can deeply analyze the browsing and purchasing behavior of users by combining user feature information with historical rating data, and recommend products that are more in line with their interests to users. In streaming services (such as Netflix and Spotify), the algorithm of the present invention can recommend film and television works and music content that meet their preferences in a targeted manner based on the user's viewing history and listening preferences. In addition, in social media platforms such as Xiaohongshu, the present invention helps users discover relevant content, potential friends and interest advertisements by analyzing the user's interests and interaction history, thereby improving user participation. In online education and news aggregation platforms, the present invention can recommend personalized courses, learning resources and news articles based on the user's learning records and reading preferences, helping users to acquire knowledge more effectively. In industries such as tourism, catering, recruitment and financial services, the present invention can also improve user experience through personalized recommendations, helping users find products or services that meet their needs. Through the combination of technologies such as clustering based on user features, personalized filling and collaborative filtering, the present invention can more accurately analyze user behavior data, thereby building an intelligent recommendation system, achieving the accuracy and diversity of recommendations, and providing innovative solutions for various industries.
[0088] The experiment of the present invention uses the public movie recommendation dataset MovieLens100K. MovieLens100K is a widely used benchmark dataset, which contains 100,000 rating records of 943 users for 1682 movies, with a rating range of 1 to 5. The dataset contains information such as the user's age, gender, occupation, and basic information about the movie. The sparsity of the dataset is 93.7%, which provides a severe testing environment for the collaborative filtering algorithm in the recommendation system. In the experimental design, 80% of the data is used as a training set and the remaining 20% of the data is used as a test set. The algorithm in this paper optimizes the performance of the collaborative filtering algorithm in a sparse matrix environment by integrating the user's personalized features (such as age, gender, occupation) with historical rating data, and verifies the effectiveness of the algorithm through experiments.
[0089] In order to measure the effect of the filling algorithm proposed in this invention, two evaluation criteria were selected in the experiment: mean absolute error (MAE) and root mean square error (RMSE). MAE and RMSE evaluate the accuracy of prediction by comparing the difference between the actual score given by the user and the score predicted by the algorithm. The formula is as follows:
[0090]
[0091] where r u,i Real rating; represents the prediction score; T is the number of samples. The lower the RMSE and MAE values, the more accurate the prediction results.
[0092] Combine the following Figure 4-5 The test of the present invention is further described.
[0093] Figure 4 A detailed comparison of the present invention (UHMF) with the recommendation algorithm (MF) processed by the majority filling method, the recommendation algorithm (UMF) processed by the user mean filling method, SPCC, OS, and BIUCF in terms of MAE (mean absolute error) is shown. The lower the MAE value, the smaller the error of the recommendation algorithm, that is, the closer the recommendation result is to the user's actual rating preference. Therefore, a lower MAE value reflects a better recommendation effect of the algorithm. The BIUCF algorithm is relatively stable in MAE performance, and the MAE value decreases slightly with the increase in the number of neighboring users. Overall, the MAE value of BIUCF is lower than that of most other algorithms, showing lower errors and higher recommendation accuracy.
[0094] Figure 5 The detailed comparison of the present invention (UHMF) with the recommendation algorithm processed by the mode filling method (MF), the recommendation algorithm processed by the user mean filling method (UMF), SPCC, OS, and BIUCF in terms of RMSE (root mean square error) is shown. The lower the RMSE value, the smaller the prediction error of the algorithm and the higher the accuracy of the recommendation. In this figure, the UHMF algorithm performs best with the lowest RMSE value, indicating that its recommendation results are closest to the user's actual rating preferences.
[0095] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.
[0096] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.
Claims
1. A hybrid filling optimization method for a recommendation system integrating user information and historical rating data, characterized in that: The following steps are involved: S1, extract and encode the user's personalized information, and then use K-means to divide the users into several clusters with similar characteristics; S2, statistically analyzes the rating behavior in each cluster, calculates the average rating of items in the cluster, and also calculates the historical average rating of each user; S3, through the regularization strategy, balances the user's historical average rating with the cluster average rating, fills in the missing values in the rating matrix, and generates a more complete rating matrix; uses the filled rating matrix to calculate the similarity, selects the K users with the highest similarity as neighbors based on the calculation results, uses the neighbor users' ratings of the target project to predict the target user's rating of the unrated project, and selects the N projects with the highest ratings as recommendations for the target user.
2. The hybrid filling optimization method for a recommendation system integrating user information and historical rating data according to claim 1, characterized in that: The S1 specifically includes: The first step is to divide the user's age into four intervals: 1-18 years old, 19-35 years old, 36-50 years old, and over 50 years old; each interval is encoded with the numbers 0, 1, 2, and 3 to form part of the age feature vector; The second step is to binary the gender feature, with male represented by 1 and female represented by 0, as an element in the feature vector; The third step is to classify occupations into four categories: professional occupations, cultural occupations, social occupations and other occupations; each category of occupation is coded with the numbers 0, 1, 2, and 3 to form part of the occupational feature vector; The fourth step is to encode the features of each user into a three-dimensional vector and use the K-means algorithm to perform clustering.
3. The hybrid filling optimization method integrating user information and historical rating data according to claim 1, characterized in that: The S2 specifically includes: In the first step, when filling in the ratings for the target users, we first find similar user clusters based on user characteristics; The second step is to count the scores of items in each cluster and calculate the average score R of the items in the cluster. c ; At the same time, a rating ratio threshold is set to ensure that only items with popularity exceeding the threshold can be used for cluster average rating calculation; in the limited item set S I and user set S U The ratio of the number of users who have rated a project to the total number of users in the project score data is the project popularity of the project, which is expressed as: Among them, if user u scores item i, then λ u,i =1, otherwise 0; |S U | represents the total number of people in the project rating data, H i Indicates the popularity of the project, and its value range is between [0,1]. i The larger the value, the more popular the project is; The third step is to take the average rating of each user on the rated items as the user's historical average rating R u .
4. The hybrid filling optimization method for a recommendation system integrating user information and historical rating data according to claim 1, characterized in that: The S3 specifically includes: In the first step, after calculating the cluster rating and user history rating, the regularization method is used to combine the two to fill the missing values in the rating matrix, which is expressed as: Where: P u,m represents the filling score of user u for item m, R u represents the average rating of user u, R c represents the average rating of item m by other users in the cluster, and represents the collective rating preference of users in the cluster; D c represents the rating deviation of cluster users on item m; λ is the regularization coefficient, which is used to balance the user's own rating preference and the collective rating within the cluster; The second step is to use the filled rating matrix to calculate the similarity. Based on the similarity calculation results, select the K users with the highest similarity as neighbors. The third step is to use the ratings of the target items by neighbor users to predict the ratings of the target users on the unrated items, which is expressed as: in The predicted score of user u for item i, is the average rating of users, r v,i The actual rating given to item i by similar user v, The average rating of similar users v, N u is a set of users similar to user u, and v∈N u is one of these similar users, sim(u,v) is the similarity between user u and user v; The fourth step is to select the N items with the highest scores according to the predicted scores as recommendations for the target users.
5. A hybrid filling optimization system for a recommendation system integrating user information and historical rating data, which implements the hybrid filling optimization method for a recommendation system integrating user information and historical rating data as claimed in any one of claims 1 to 4, characterized in that: include: The user clustering module extracts and encodes the user's personalized information, and then uses K-means to divide the user into several clusters with similar characteristics; The historical average rating calculation module performs statistical analysis on the rating behavior in each cluster, calculates the average rating of items in the cluster, and also calculates the historical average rating of each user; The target user recommendation module uses a regularization strategy to balance the user's historical average rating with the cluster average rating, fill in the missing values in the rating matrix, and generate a more complete rating matrix; the similarity is calculated using the filled rating matrix, and the K users with the highest similarity are selected as neighbors based on the calculation results. The ratings of the target items by the neighbor users are used to predict the target user's ratings for the unrated items, and the N items with the highest ratings are selected as recommendations for the target user.
6. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the hybrid filling optimization method that integrates user information and historical rating data as described in any one of claims 1 to 4.
7. A computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the hybrid filling optimization method for integrating user information and historical rating data as described in any one of claims 1 to 4.
8. An information data processing terminal, comprising the recommendation system hybrid filling optimization system integrating user information and historical rating data as claimed in claim 5.