Method and system for recommending goods based on audience characteristics

Abstract

The application discloses a commodity recommendation method and system based on audience characteristics, and relates to the technical field of computers, and the method comprises the following steps: collecting real-time viewing data of users from different viewing platforms; extracting user characteristics, film characteristics and viewing interaction characteristics from the collected real-time viewing data of the users; wherein the viewing interaction characteristics comprise viewing behavior characteristics, scoring and feedback characteristics, preference and interest characteristics and social interaction characteristics; obtaining to-be-recommended resources in a commodity recommendation resource pool; extracting commodity characteristics of the to-be-recommended resources; and updating the to-be-recommended resources in the commodity recommendation resource pool according to the commodity characteristics of the to-be-recommended resources, the user characteristics, the film characteristics and the viewing interaction characteristics.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more specifically, to a method and system for recommending products based on the characteristics of moviegoers. Background Technology

[0002] With the widespread adoption of the internet and the rapid development of e-commerce, users are faced with a vast array of products to choose from. Finding items that match their needs and interests from this vast selection has become a challenge. Recommendation systems have emerged to address this need. By analyzing user preferences, they recommend products that users may be interested in, thereby improving the user shopping experience and the platform's operational efficiency.

[0003] However, most existing product recommendation systems are based on general user behavior data and do not fully consider the specific characteristics of moviegoers. User preferences may change with the release of new films, and existing product recommendation systems cannot adapt to these changes in a timely manner, leading to a decline in the quality of recommendation results.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a product recommendation method and system based on moviegoer characteristics to solve the above-mentioned technical problems.

[0006] This application provides a product recommendation method based on movie-watching audience characteristics, including: collecting real-time movie-watching data from users on different movie-watching platforms; extracting user characteristics, film characteristics, and movie-watching interaction characteristics from the collected real-time movie-watching data; wherein, the movie-watching interaction characteristics include movie-watching behavior characteristics, rating and feedback characteristics, preference and interest characteristics, and social interaction characteristics; obtaining resources to be recommended from a product recommendation resource pool; extracting product characteristics from the resources to be recommended; and updating the resources to be recommended in the product recommendation resource pool according to the product characteristics, user characteristics, film characteristics, and movie-watching interaction characteristics.

[0007] This application provides a product recommendation system based on movie-watching audience characteristics, comprising: a movie-watching data collection module for collecting real-time movie-watching data from users on different movie-watching platforms; a movie-watching feature extraction module for extracting user features, movie features, and movie-watching interaction features from the collected real-time movie-watching data; wherein the movie-watching interaction features include movie-watching behavior features, rating and feedback features, preference and interest features, and social interaction features; a resource to be recommended module for acquiring resources to be recommended from a product recommendation resource pool; a product feature extraction module for extracting product features from the resources to be recommended; and a resource to be recommended update module for updating the resources to be recommended in the product recommendation resource pool based on the product features, user features, movie features, and movie-watching interaction features.

[0008] Furthermore, the product recommendation system based on moviegoer characteristics also includes:

[0009] The resource classification and tag generation module is used to classify the resources to be recommended in the product recommendation resource pool according to the product characteristics, product type, brand, price range and applicable scenario dimensions, so as to generate product tag information matching the resources to be recommended.

[0010] Furthermore, the movie-watching feature extraction module extracts user features and movie features from the collected real-time movie-watching data of users, and is configured as follows:

[0011] An autoencoder is used to encode the user's real-time viewing data to map the user's real-time viewing data into an embedding space, thereby obtaining the user's embedding representation; wherein, the number of features contained in the user's embedding representation is less than the number of features in the user's real-time viewing data.

[0012] The user characteristics are determined based on the user's embedded representation;

[0013] By utilizing a multimodal video feature extraction framework, sentiment analysis and topic mining are performed on users' real-time viewing data to construct a film knowledge graph;

[0014] The film knowledge graph is analyzed to obtain the film features; wherein, the film features include the film's creation and performance features, visual appearance features, optical flow features and audio features; the creation and performance features include the features of the creator, the features of the performer, and the content classification features.

[0015] Furthermore, the movie-watching feature extraction module extracts movie-watching interaction features from the collected real-time movie-watching data of users, and is configured as follows:

[0016] Recurrent neural networks are used to model the viewing behavior sequence data in the user's real-time viewing data. Combined with scene perception algorithms, the user's viewing device, time, and location are incorporated into the viewing behavior sequence data.

[0017] Based on the movie-watching behavior sequence data, determine the association between the movie-watching behavior set and the movie-watching behavior;

[0018] Each movie-watching behavior in the set of movie-watching behaviors is treated as a node, and the edges between nodes represent the association weights between the movie-watching behaviors, so as to construct a user movie-watching behavior graph.

[0019] The user movie-watching behavior map is analyzed by graph neural network to uncover potential patterns and rules of user movie-watching behavior and obtain movie-watching behavior characteristics;

[0020] Perform text sentiment analysis on the rating feedback data in users' real-time movie viewing data to extract the text sentiment features and the voice sentiment features of users when rating.

[0021] The text sentiment features and the voice sentiment features are fused to obtain the rating and feedback features;

[0022] By combining online learning algorithms with time series analysis algorithms, user preference features are extracted from real-time viewing data, and a user interest evolution graph is constructed. The interest evolution graph is used to record the user's current preferences and interest points, as well as to trace the source and development path of interests.

[0023] Based on the interest evolution graph, the interest associations of users in different fields are mined to obtain the preferences and interest features;

[0024] Construct a dynamic user-viewing social network; wherein, the user-viewing social network is used to represent users' social relationships and interactive behaviors;

[0025] Based on the aforementioned user movie-watching social network, and combined with the influence propagation model, we analyze the user's influence, emotional tendencies, and information propagation paths within the social network.

[0026] The social interaction characteristics are determined based on the user's influence, emotional inclination, and information dissemination path on the social network.

[0027] Furthermore, the resource update module updates the resources to be recommended in the product recommendation resource pool based on the product characteristics, user characteristics, film characteristics, and viewing interaction characteristics of the resources to be recommended, and is configured as follows:

[0028] The missing values ​​of the user features, the missing values ​​of the film features, and the missing values ​​of the viewing interaction features are filled using a K-means clustering algorithm.

[0029] Specifically, for categorical features among the user features, film features, and viewing interaction features, adaptive encoding combined with leave-one-out encoding is used for processing; for continuous features among the user features, film features, and viewing interaction features, a custom quantile Min-Max standardization is used for processing, first dividing the features into segments according to quantiles, and then performing Min-Max standardization within each segment.

[0030] Based on the augmented viewing interaction features, augmented user features, and augmented film features, a user preference type score feature and a weighted feature of viewing popularity rating are derived using a word frequency-inverse document frequency algorithm; wherein, the weighted feature of viewing popularity rating is obtained by weighted fusion of the augmented film features and the user preference type score feature.

[0031] The imputed viewing interaction features, imputed user features, and imputed film features are subjected to singular value decomposition to obtain singular values ​​and singular vectors.

[0032] The singular vectors corresponding to the first y1 largest singular values ​​are selected as the feature vectors after dimensionality reduction; where y1 is a natural number, and the value of y1 is determined by the cumulative explained variance ratio.

[0033] The augmented viewing interaction features, augmented user features, and augmented film features are projected onto the dimension-reduced feature vector to obtain the dimension-reduced data representation.

[0034] Based on the dimensionality-reduced data representation, an intra-class scatter matrix and an inter-class scatter matrix are constructed; wherein, the intra-class scatter matrix is ​​used to represent the scatter of samples within the same class; and the inter-class scatter matrix is ​​used to represent the scatter of samples between different classes.

[0035] The projection matrix is ​​obtained by multiplying the inverse of the intra-class scatter matrix with the inter-class scatter matrix.

[0036] Calculate the eigenvalues ​​and eigenvectors of the projection matrix; wherein the eigenvalues ​​of the projection matrix represent the ability of the eigenvectors of the projection matrix to distinguish categories;

[0037] Based on the user preference type score features and the weighted features of movie popularity scores, the first y2 generalized feature vectors are selected from the feature vectors of the projection matrix; where y2 is a natural number.

[0038] The filled-in movie-watching interaction features, filled-in user features, and filled-in movie features are projected onto the generalized feature vector to obtain the projected feature representation;

[0039] The projection feature representation is evaluated and optimized to obtain optimized projection viewing features;

[0040] The resources to be recommended in the product recommendation resource pool are updated based on the product tag information that matches the resources to be recommended and the optimized projection viewing features.

[0041] Furthermore, the evaluation and optimization of the projection feature representation to obtain optimized projection viewing features is configured as follows:

[0042] An F-test is performed on the projected feature representation to process features that do not conform to a normal distribution, and features with significant F-statistics are reconstructed.

[0043] Using gradient boosting tree as the base model, recursive feature elimination is performed on the projected features after F test, eliminating the 15% of features with the smallest weight coefficient in each round.

[0044] The projected features after recursive feature elimination were trained using the XGBoost model and the LightGBM model. The first importance of the projected features after recursive feature elimination in the XGBoost model and the second importance in the LightGBM model were calculated respectively.

[0045] The first and second importance of the projected features after recursive feature elimination are weighted and fused to obtain the reference importance of the projected features after recursive feature elimination.

[0046] The top 80% of features by reference importance are retained to form a feature candidate set;

[0047] Based on the data distribution of the feature candidate set, adaptively select 7-fold cross-validation or 5-fold cross-validation;

[0048] The feature candidate set is evaluated using 7-fold cross-validation or 5-fold cross-validation combined with hierarchical sampling;

[0049] Based on the evaluation results of the feature candidate set, an adaptive sensitivity analysis is performed on each feature in the feature candidate set using a feature contribution analysis algorithm based on Shapley value and LIME algorithm, and a sensitivity threshold is dynamically set according to the importance and relevance of the feature.

[0050] For features below the sensitivity threshold, feature pruning is performed;

[0051] For the pruned feature candidate set, an adaptive feature fusion algorithm based on deep learning attention mechanism is used to fuse the features in the pruned feature candidate set to obtain the optimized projection viewing features.

[0052] Furthermore, the intra-class scatter matrix is ​​S ω The inter-class scatter matrix is ​​S. c The projection matrix is ​​inv(S) ω )×S c The calculation process of the projection matrix is ​​as follows:

[0053] Calculate S ω inverse matrix inv(S) ω );

[0054] inv(S) ω ) and S c Multiplying them together, we get inv(S) ω )×S c ;

[0055] Wherein, the inverse matrix inv(S) ω ) is a square matrix; inv(S ω )×S c The feature vectors are used to maximize inter-class separation and minimize intra-class separation.

[0056] Furthermore, the movie viewing data collection module collects real-time movie viewing data from users on different movie viewing platforms, and is configured to: collect data from cinema ticketing systems, online video platform viewing records, and movie-related discussion data on social media; and integrate the collected data from cinema ticketing systems, online video platform viewing records, and movie-related discussion data on social media into real-time movie viewing data for users.

[0057] Based on the embodiments provided in this application, a movie viewing data collection module collects user movie viewing data in real time across different movie viewing platforms. This allows for the timely acquisition of multi-dimensional information such as user viewing behavior, rating feedback, preferences, interests, and social interactions related to new films. This real-time data collection method enables the system to keenly perceive subtle changes in the preferences of moviegoers, such as shifts in interest due to the release of new films. This provides the recommendation system with the latest and most accurate basis for user preferences, effectively solving the problem that existing recommendation systems cannot adapt to changes in moviegoer preferences in a timely manner. The movie viewing feature extraction module comprehensively extracts user features, film features, and movie viewing interaction features. Movie viewing interaction features encompass multiple aspects, including movie viewing behavior features, rating and feedback features, preference and interest features, and social interaction features. This comprehensive and multi-layered feature extraction method can fully characterize users' movie viewing habits and preferences, as well as the complex interactive relationship between users and films. Compared to recommendation systems based solely on general user behavior data, the system in this application can more accurately understand users' personalized needs, thereby providing product recommendations that better match users' current interests and preferences, significantly improving the quality of recommendation results and user satisfaction. The resource update module dynamically updates the resources in the product recommendation resource pool based on extracted features. This mechanism ensures that recommended resources always match the user's latest viewing characteristics and preferences, promptly introducing new products related to the user's current viewing interests and eliminating old products that no longer meet the user's needs. Through this dynamic update, the recommendation system can continuously provide users with fresh and accurate recommendations, avoiding lag and homogenization of recommendation results, further improving the user's shopping experience and the platform's operational efficiency. The recommendation system in this application is specifically designed for moviegoers, fully considering the special characteristics and needs of this group. Compared with general recommendation systems, it can better adapt to the diverse and personalized characteristics of moviegoers, providing customized recommendation services for users with different viewing preferences and viewing frequencies. Attached Figure Description

[0058] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0059] Figure 1 This is a flowchart of an optional product recommendation method based on moviegoer characteristics according to an embodiment of this application;

[0060] Figure 2 This is a structural diagram of an optional product recommendation system based on moviegoer characteristics according to an embodiment of this application.

[0061] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0062] 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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0063] Optionally, such as Figure 1 As shown, this application provides a product recommendation method based on the characteristics of moviegoers, including:

[0064] S101 collects real-time viewing data from users on different viewing platforms;

[0065] S102, extract user features, film features, and viewing interaction features from the collected real-time viewing data of users; among which, viewing interaction features include viewing behavior features, rating and feedback features, preference and interest features, and social interaction features;

[0066] In this embodiment, the characteristics of movie-watching behavior include:

[0067] Viewing frequency: The number of times a user watches a movie within a specific time period, reflecting the user's movie-watching activity, such as the number of movies watched per month;

[0068] Viewing duration: The average length of time a user watches each movie, reflecting their level of focus. For example, the average length of each viewing session;

[0069] Viewing time: The preferred time of day for users to watch movies, such as weekends, evenings, and holidays, which reflects users' viewing habits. For example, users mainly watch movies between 8 pm and 10 pm.

[0070] Viewing device: The type of device a user uses to watch movies, such as a mobile phone, tablet, computer, or TV. This reflects the user's viewing device preference; for example, the user mainly uses a mobile phone to watch movies.

[0071] Rating and feedback features include:

[0072] Rating distribution: User ratings for different movies reflect the strength of user preferences. For example, science fiction movies generally receive higher ratings.

[0073] Comment content: User comments on movies reflect their detailed preferences and emotional inclinations. For example, a user might mention in their comment that they "liked the special effects in the movie" or "disliked the plot."

[0074] Likes and shares: Users' liking and sharing behavior of movies reflects their recommendation intentions. For example, users frequently like and share their favorite movies to social media;

[0075] Preference and interest characteristics include:

[0076] Movie Genre Preferences: User's preferred movie genres, such as science fiction, action, romance, comedy, etc., reflect their areas of interest. For example, users primarily watch science fiction and action movies;

[0077] Director and actor preferences: User preferences for directors and actors reflect specific user preferences. For example, a user might like movies directed by Christopher Nolan or starring Tom Cruise;

[0078] Theme and genre preferences: Users' preferred film themes and genres, such as science fiction adventure, historical war, and love stories, reflect their deeper interests. For example, users may enjoy war movies with historical backgrounds;

[0079] Social interaction characteristics include:

[0080] Social Network Interaction: User interactions related to the film on social networks, such as following official film accounts and participating in film-related discussions, reflect users' social preferences. For example, users frequently participate in film-related Weibo discussions.

[0081] Friend recommendations: A user's behavior of watching movies based on friend recommendations reflects their social influence. For example, a user frequently watches movies based on their friends' recommendations;

[0082] Audience demographics: This reflects the user's social circle and movie-watching habits, indicating who they frequently watch movies with. For example, users often watch movies with family or friends.

[0083] S103, Obtain the recommended resources from the product recommendation resource pool; Extract the product features of the recommended resources;

[0084] S104. Update the resources to be recommended in the product recommendation resource pool based on the product characteristics, user characteristics, film characteristics, and viewing interaction characteristics of the resources to be recommended.

[0085] Based on the embodiments provided in this application, a movie viewing data collection module collects user movie viewing data in real time across different movie viewing platforms. This allows for the timely acquisition of multi-dimensional information such as user viewing behavior, rating feedback, preferences, interests, and social interactions related to new films. This real-time data collection method enables the system to keenly perceive subtle changes in the preferences of moviegoers, such as shifts in interest due to the release of new films. This provides the recommendation system with the latest and most accurate basis for user preferences, effectively solving the problem that existing recommendation systems cannot adapt to changes in moviegoer preferences in a timely manner. The movie viewing feature extraction module comprehensively extracts user features, film features, and movie viewing interaction features. Movie viewing interaction features encompass multiple aspects, including movie viewing behavior features, rating and feedback features, preference and interest features, and social interaction features. This comprehensive and multi-layered feature extraction method can fully characterize users' movie viewing habits and preferences, as well as the complex interactive relationship between users and films. Compared to recommendation systems based solely on general user behavior data, the system in this application can more accurately understand users' personalized needs, thereby providing product recommendations that better match users' current interests and preferences, significantly improving the quality of recommendation results and user satisfaction. The resource update module dynamically updates the resources in the product recommendation resource pool based on extracted features. This mechanism ensures that recommended resources always match the user's latest viewing characteristics and preferences, promptly introducing new products related to the user's current viewing interests and eliminating old products that no longer meet the user's needs. Through this dynamic update, the recommendation system can continuously provide users with fresh and accurate recommendations, avoiding lag and homogenization of recommendation results, further improving the user's shopping experience and the platform's operational efficiency. The recommendation system in this application is specifically designed for moviegoers, fully considering the special characteristics and needs of this group. Compared with general recommendation systems, it can better adapt to the diverse and personalized characteristics of moviegoers, providing customized recommendation services for users with different viewing preferences and viewing frequencies.

[0086] Optionally, such as Figure 2 As shown, this application provides a product recommendation system based on moviegoer characteristics, including:

[0087] The movie viewing data collection module 201 is used to collect real-time movie viewing data from users on different movie viewing platforms;

[0088] The movie viewing feature extraction module 202 is used to extract user features, movie features, and movie viewing interaction features from the collected real-time movie viewing data of users; among which, movie viewing interaction features include movie viewing behavior features, rating and feedback features, preference and interest features, and social interaction features.

[0089] The module 203 for acquiring resources to be recommended is used to acquire resources to be recommended from the product recommendation resource pool.

[0090] The product feature extraction module 204 is used to extract product features of the resources to be recommended;

[0091] The resource update module 205 is used to update the resources to be recommended in the product recommendation resource pool based on the product characteristics, user characteristics, film characteristics, and viewing interaction characteristics of the resources to be recommended.

[0092] Furthermore, product recommendation systems based on moviegoer characteristics also include:

[0093] The resource classification and tag generation module is used to classify the resources to be recommended in the product recommendation resource pool according to the product characteristics, product type, brand, price range and applicable scenario, so as to generate product tag information that matches the resources to be recommended.

[0094] Furthermore, the movie viewing feature extraction module extracts user features and movie features from the collected real-time movie viewing data of users, and is configured as follows:

[0095] An autoencoder is used to encode the user's real-time viewing data to map the user's real-time viewing data into the embedding space, thereby obtaining the user's embedding representation and achieving data dimensionality reduction; wherein, the number of features contained in the user's embedding representation is less than the number of features in the user's real-time viewing data.

[0096] Determine user characteristics based on the user's embedded representation;

[0097] A multimodal video feature extraction framework is used to perform sentiment analysis and topic mining on users' real-time viewing data in order to construct a film knowledge graph; the multimodal video feature extraction framework may include the video_features framework.

[0098] In this embodiment, sentiment analysis and topic modeling are performed on text content such as movie synopses and reviews to extract useful keywords and sentiment information. The Transformer model in deep learning can be used for topic mining, which can more deeply understand the keywords and the sentiment relationships between them in user feedback text.

[0099] The film knowledge graph was analyzed to obtain film features, which include creative and expressive features, visual appearance features, optical flow features, and audio features. Creative and expressive features include features of the creative subject, features of the performing subject, and content classification features.

[0100] Furthermore, the viewing feature extraction module extracts viewing interaction features from the collected real-time viewing data of users, and is configured as follows:

[0101] Recurrent neural networks are used to model the viewing behavior sequence data in the real-time viewing data of users. Combined with scene perception algorithms, the user's viewing device, time and location are integrated into the viewing behavior sequence data.

[0102] Based on movie-watching behavior sequence data, determine the relationship between movie-watching behavior sets and movie-watching behaviors; among which, movie-watching behavior relationships include continuous viewing, intermittent viewing, etc.

[0103] Each movie-watching behavior in the movie-watching behavior set is treated as a node, and the edges between nodes represent the association weights between the movie-watching behaviors, in order to construct a user movie-watching behavior graph;

[0104] The correlation weights between various movie-watching behaviors are determined based on the following formula:

[0105]

[0106] Among them, W ij This represents the association weight between viewing behavior i and viewing behavior j; v i and v j These are the vector representations of viewing behaviors i and j in the embedding space, respectively; cos(v i ,v j ) is a vector v i sum vector v j The cosine similarity is used to measure the similarity of viewing behaviors in terms of content features; N ij N is the number of times that movie-watching behavior i and movie-watching behavior j occur simultaneously; i and N j α and β are the total number of times viewing behavior i and viewing behavior j occur, respectively; α and β are hyperparameters used to balance the influence of content similarity and co-occurrence frequency on association weights.

[0107] By analyzing the user's movie-watching behavior graph using graph neural networks, we can uncover the potential patterns and rules of user movie-watching behavior and obtain movie-watching behavior characteristics.

[0108] Perform text sentiment analysis on the rating feedback data in users' real-time movie viewing data to extract the text sentiment features and the voice sentiment features of users when rating.

[0109] Based on the following formula, text sentiment features and speech sentiment features are fused to obtain rating and feedback features;

[0110] f = tanh(W) t t+W v v+b)⊙m

[0111] Where f represents the rating and feedback features; W t t represents the sentiment feature of the text, weighted by the weight matrix W.t Perform a linear transformation; W v v represents the emotional features of the speech, and v is weighted by the matrix W. v A linear transformation is performed; b is the bias vector used to adjust the translation of the fused features; tanh is the hyperbolic tangent activation function used to introduce nonlinearity so that the fused features can better capture complex emotional information; ⊙ is element-wise multiplication; m is the modality importance vector used to adjust the contribution of different modal features.

[0112] By combining online learning algorithms with time series analysis algorithms, user preference features are extracted from real-time viewing data, and a user interest evolution graph is constructed. The interest evolution graph is used to record the user's current preferences and interest points, as well as to trace the source and development path of interests.

[0113] Based on the interest evolution map, the system can explore the interest associations between users in different fields to obtain preferences and interest characteristics. For example, if the system finds that a user likes the soundtrack of a movie, it will automatically recommend other movies with similar music styles to the user, thus realizing cross-field interest expansion and recommendation.

[0114] Construct a dynamic user-viewing social network; whereby the user-viewing social network is used to represent users' social relationships and interactive behaviors;

[0115] Based on users' movie-watching social networks, and combined with an influence propagation model, this study analyzes users' influence, sentiment, and information dissemination paths on social networks. The influence propagation model may include, but is not limited to, the independent cascade model and the linear threshold model.

[0116] Social interaction characteristics are determined based on users' influence, emotional inclinations, and information dissemination paths on social networks.

[0117] Furthermore, the resource update module updates the resources to be recommended in the product recommendation resource pool based on the product characteristics, user characteristics, film characteristics, and viewing interaction characteristics of the resources to be recommended, and is configured as follows:

[0118] The missing values ​​of user features, film features, and viewing interaction features were filled using a K-means clustering algorithm.

[0119] Among them, for categorical features in user features, film features, and viewing interaction features, adaptive coding is used in combination with leave-one-out coding; for continuous features in user features, film features, and viewing interaction features, custom quantile Min-Max standardization is used, which first segments the features according to quantiles, and then performs Min-Max standardization within each segment.

[0120] Among them, the custom quantile Min-Max standardization is a data standardization method that combines the concept of quantiles with the traditional Min-Max standardization method;

[0121] Based on the imputed viewing interaction features, imputed user features, and imputed film features, user preference type score features and viewing popularity score weighted features are derived through the word frequency-inverse document frequency algorithm; among them, the viewing popularity score weighted features are obtained by weighted fusion of the imputed film features and user preference type score features.

[0122] The imputed viewing interaction features, imputed user features, and imputed film features are subjected to singular value decomposition to obtain singular values ​​and singular vectors.

[0123] The singular vectors corresponding to the first y1 largest singular values ​​are selected as the feature vectors after dimensionality reduction; where y1 is a natural number, and the value of y1 is determined by the cumulative explained variance ratio. Usually, the number of singular values ​​with a cumulative explained variance ratio of 85%-90% is selected.

[0124] The augmented viewing interaction features, augmented user features, and augmented film features are projected onto the dimension-reduced feature vector to obtain the dimension-reduced data representation.

[0125] Based on the dimensionality-reduced data representation, we construct intra-class scatter matrices and inter-class scatter matrices; the intra-class scatter matrix is ​​used to represent the scatter of samples within the same class; the inter-class scatter matrix is ​​used to represent the scatter of samples between different classes.

[0126] The projection matrix is ​​obtained by multiplying the inverse of the intra-class scatter matrix with the inter-class scatter matrix.

[0127] Calculate the eigenvalues ​​and eigenvectors of the projection matrix; where the eigenvalues ​​of the projection matrix represent the ability of the eigenvectors of the projection matrix to distinguish between categories.

[0128] Based on the user preference type score features and the weighted features of movie popularity scores, the first y2 generalized feature vectors are selected from the feature vectors of the projection matrix; where y2 is a natural number, usually less than the number of categories minus one. These feature vectors constitute the projection matrix, which can maximize the difference between categories and minimize the difference within categories.

[0129] In this embodiment, the specific steps for selecting the generalized feature vector may include:

[0130] Constructing a feature matrix: Integrate the imputed viewing interaction features, imputed user features, and imputed film features into a single feature matrix. Each row of this matrix represents a user or a film, and each column represents a feature.

[0131] Applying Singular Value Decomposition (SVD): Perform SVD on the eigenvalue matrix. The decomposition yields three matrices: one diagonal matrix containing the singular values, and the other two orthogonal matrices. The values ​​on the diagonal of the singular value matrix represent the importance of each eigenvector.

[0132] Selecting eigenvectors: Choose the top y2 largest singular values ​​from the singular value matrix. The eigenvectors corresponding to these singular values ​​are the most important generalized eigenvectors. Specifically, select the eigenvectors with the largest y2 singular values. These eigenvectors can capture the main changes and trends in the data, and are therefore called "generalized eigenvectors".

[0133] Determine the value of y2: y2 is a natural number representing the number of feature vectors selected. The value of y2 can be determined based on the specific application scenario and requirements. Generally, the smaller the value of y2, the better the feature vectors capture the main information in the data, but some details may be lost; the larger the value of y2, the more details the feature vectors capture, but the computational complexity will also increase.

[0134] By following the steps above, the top y2 most important generalized feature vectors can be selected from the feature matrix. These feature vectors can effectively represent the user preference type score features and the movie popularity score weighted features, thereby improving the accuracy and relevance of the recommendation system.

[0135] The filled-in movie-watching interaction features, filled-in user features, and filled-in movie features are projected onto the generalized feature vector to obtain the projected feature representation;

[0136] The projection feature representation is evaluated and optimized to obtain optimized projection viewing features;

[0137] Based on the product tag information that matches the resources to be recommended and the optimized projection viewing characteristics, the resources to be recommended in the product recommendation resource pool are updated.

[0138] Furthermore, the projection feature representation is evaluated and optimized to obtain optimized projection viewing features, which are configured as follows:

[0139] Perform an F-test on the projected feature representation, handle features that do not conform to the normal distribution, and reconstruct features with significant F-statistics.

[0140] The F-test is a statistical test used to compare two or more variances to determine whether there are significant differences between them. In practice, the F-test is commonly used in analysis of variance and regression analysis. Specifically, the F-test determines whether the hypothesis is true by calculating the F-statistic. The F-statistic is obtained by comparing the ratio of the between-group mean square to the within-group mean square. If the F-statistic is large, the corresponding p-value is small, indicating that at least one coefficient is significantly different from zero, and the model as a whole is meaningful.

[0141] The F-statistic is used to measure the significance of a model or feature. A larger F-statistic indicates stronger overall significance and a better model fit. In an F-test, if the calculated F-statistic is greater than the critical value at a given significance level, the null hypothesis can be rejected, and the model or feature is considered significant.

[0142] Using gradient boosting trees as the base model, recursive feature elimination is performed on the projected features after the F test. In each round, the 15% of features with the smallest weight coefficients are eliminated. For example, if the current feature set has 100 features, then in each round, 15 features with the smallest weight coefficients will be eliminated, and 85 features will be retained for the next round of training.

[0143] The projected features after recursive feature elimination were trained using the XGBoost model and the LightGBM model. The first importance of the projected features after recursive feature elimination in the XGBoost model and the second importance in the LightGBM model were calculated respectively.

[0144] The first and second importance of the projected features after recursive feature elimination are weighted and fused to obtain the reference importance of the projected features after recursive feature elimination.

[0145] The top 80% of features by reference importance are retained to form a feature candidate set;

[0146] Based on the data distribution of the feature candidate set, adaptively select 7-fold cross-validation or 5-fold cross-validation;

[0147] The feature candidate set is evaluated using 7-fold or 5-fold cross-validation combined with stratified sampling. For example, a custom comprehensive evaluation metric can be used to evaluate the feature candidate set, which integrates accuracy, recall, F1-score, and ROC-AUC. The weights of each metric are dynamically adjusted according to the needs of different business scenarios to comprehensively evaluate the final feature candidate set.

[0148] Based on the evaluation results of the feature candidate set, an adaptive sensitivity analysis is performed on each feature in the feature candidate set using a feature contribution analysis algorithm based on Shapley value and LIME algorithm. The sensitivity threshold is dynamically set according to the importance and relevance of the feature.

[0149] For features below the sensitivity threshold, feature pruning is performed;

[0150] For the pruned feature candidate set, an adaptive feature fusion algorithm based on deep learning attention mechanism is used to fuse the features in the pruned feature candidate set to obtain optimized projection viewing features.

[0151] Furthermore, the intra-class scatter matrix is ​​S ω The inter-class scatter matrix is ​​S. c The projection matrix is ​​inv(S) ω )×S c The calculation process of the projection matrix is ​​as follows:

[0152] Calculate S ω inverse matrix inv(S) ω );

[0153] inv(S) ω ) and S c Multiplying them together, we get inv(S) ω )×S c ;

[0154] Wherein, the inverse matrix inv(S) ω ) is a square matrix; inv(S ω )×S c The feature vectors are used to maximize inter-class separation and minimize intra-class separation.

[0155] S ω =∑ x∈categ N categ (x-μ categ (x-μ) categ ) T

[0156] S c =∑ x∈categ N categ (μ categ -μ)(μ categ -μ) T

[0157] Where x is a sample in category categ; μ categ It is the mean vector of category; T is the matrix transpose; N categ is the number of samples in category 'categ'; μ is the population mean vector of all samples.

[0158] Furthermore, the movie viewing data collection module collects real-time movie viewing data from users on different movie viewing platforms, and is configured as follows:

[0159] Collect data from cinema ticketing systems, online video platform viewing records, and movie-related discussions on social media.

[0160] The collected data from cinema ticketing systems, online video platform viewing records, and movie-related discussions on social media are integrated into real-time movie-watching data for users.

[0161] It should be noted that the embodiments implemented on the product recommendation system side based on moviegoer characteristics in this application can be referenced with the embodiments implemented on the product recommendation method side based on moviegoer characteristics, and will not be described in detail here.

[0162] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A product recommendation method based on moviegoer characteristics, characterized in that, include: Collect real-time viewing data from users on different viewing platforms; User features, film features, and viewing interaction features are extracted from the collected real-time viewing data of users; wherein, the viewing interaction features include viewing behavior features, rating and feedback features, preference and interest features, and social interaction features; Obtain the recommended resources from the product recommendation resource pool; extract the product features of the recommended resources; The resources to be recommended in the product recommendation resource pool are updated based on the product characteristics, user characteristics, film characteristics, and viewing interaction characteristics of the resources to be recommended. The method further includes: classifying the resources to be recommended according to the product characteristics of the resources to be recommended in the product recommendation resource pool, based on the product type dimension, brand dimension, price range dimension, and applicable scenario dimension, so as to generate product tag information matching the resources to be recommended; The step of updating the resources to be recommended in the product recommendation resource pool based on the product characteristics, user characteristics, film characteristics, and viewing interaction characteristics of the resources to be recommended is configured as follows: The missing values ​​of the user features, the missing values ​​of the film features, and the missing values ​​of the viewing interaction features are filled using a K-means clustering algorithm. Specifically, for categorical features among the user features, film features, and viewing interaction features, adaptive encoding combined with leave-one-out encoding is used for processing; for continuous features among the user features, film features, and viewing interaction features, a custom quantile Min-Max standardization is used for processing, first dividing the features into segments according to quantiles, and then performing Min-Max standardization within each segment. Based on the augmented viewing interaction features, augmented user features, and augmented film features, a user preference type score feature and a weighted feature of viewing popularity rating are derived using a word frequency-inverse document frequency algorithm; wherein, the weighted feature of viewing popularity rating is obtained by weighted fusion of the augmented film features and the user preference type score feature. The imputed viewing interaction features, imputed user features, and imputed film features are subjected to singular value decomposition to obtain singular values ​​and singular vectors. The singular vectors corresponding to the first y1 largest singular values ​​are selected as the feature vectors after dimensionality reduction; where y1 is a natural number, and the value of y1 is determined by the cumulative explained variance ratio. The augmented viewing interaction features, augmented user features, and augmented film features are projected onto the dimension-reduced feature vector to obtain the dimension-reduced data representation. Based on the dimensionality-reduced data representation, an intra-class scatter matrix and an inter-class scatter matrix are constructed; wherein, the intra-class scatter matrix is ​​used to represent the scatter of samples within the same class; and the inter-class scatter matrix is ​​used to represent the scatter of samples between different classes. The projection matrix is ​​obtained by multiplying the inverse of the intra-class scatter matrix with the inter-class scatter matrix. Calculate the eigenvalues ​​and eigenvectors of the projection matrix; wherein the eigenvalues ​​of the projection matrix represent the ability of the eigenvectors of the projection matrix to distinguish categories; Based on the user preference type score features and the weighted features of movie popularity scores, the first y2 generalized feature vectors are selected from the feature vectors of the projection matrix; where y2 is a natural number. The filled-in movie-watching interaction features, filled-in user features, and filled-in movie features are projected onto the generalized feature vector to obtain the projected feature representation; The projection feature representation is evaluated and optimized to obtain optimized projection viewing features; The resources to be recommended in the product recommendation resource pool are updated based on the product tag information that matches the resources to be recommended and the optimized projection viewing features.

2. A product recommendation system based on moviegoer characteristics, wherein the system implements the method as described in claim 1, characterized in that, include: The movie viewing data collection module is used to collect real-time movie viewing data from users on different movie viewing platforms; The movie viewing feature extraction module is used to extract user features, movie features, and movie viewing interaction features from the collected real-time movie viewing data of users; wherein, the movie viewing interaction features include movie viewing behavior features, rating and feedback features, preference and interest features, and social interaction features; The module for acquiring resources to be recommended is used to acquire resources to be recommended from the product recommendation resource pool. The product feature extraction module is used to extract the product features of the resource to be recommended; The resource update module is used to update the resources to be recommended in the product recommendation resource pool based on the product characteristics, user characteristics, film characteristics, and viewing interaction characteristics of the resources to be recommended. The product recommendation system based on moviegoer characteristics also includes: a resource classification and tag generation module, which is used to classify the resources to be recommended in the product recommendation resource pool according to the product characteristics, product type dimension, brand dimension, price range dimension and applicable scenario dimension, so as to generate product tag information matching the resources to be recommended; The resource update module updates the resources to be recommended in the product recommendation resource pool based on the product characteristics, user characteristics, film characteristics, and viewing interaction characteristics of the resources to be recommended, and is configured as follows: The missing values ​​of the user features, the missing values ​​of the film features, and the missing values ​​of the viewing interaction features are filled using a K-means clustering algorithm. Specifically, for categorical features among the user features, film features, and viewing interaction features, adaptive encoding combined with leave-one-out encoding is used for processing; for continuous features among the user features, film features, and viewing interaction features, a custom quantile Min-Max standardization is used for processing, first dividing the features into segments according to quantiles, and then performing Min-Max standardization within each segment. Based on the augmented viewing interaction features, augmented user features, and augmented film features, a user preference type score feature and a weighted feature of viewing popularity rating are derived using a word frequency-inverse document frequency algorithm; wherein, the weighted feature of viewing popularity rating is obtained by weighted fusion of the augmented film features and the user preference type score feature. The imputed viewing interaction features, imputed user features, and imputed film features are subjected to singular value decomposition to obtain singular values ​​and singular vectors. The singular vectors corresponding to the first y1 largest singular values ​​are selected as the feature vectors after dimensionality reduction; where y1 is a natural number, and the value of y1 is determined by the cumulative explained variance ratio. The augmented viewing interaction features, augmented user features, and augmented film features are projected onto the dimension-reduced feature vector to obtain the dimension-reduced data representation. Based on the dimensionality-reduced data representation, an intra-class scatter matrix and an inter-class scatter matrix are constructed; wherein, the intra-class scatter matrix is ​​used to represent the scatter of samples within the same class; and the inter-class scatter matrix is ​​used to represent the scatter of samples between different classes. The projection matrix is ​​obtained by multiplying the inverse of the intra-class scatter matrix with the inter-class scatter matrix. Calculate the eigenvalues ​​and eigenvectors of the projection matrix; wherein the eigenvalues ​​of the projection matrix represent the ability of the eigenvectors of the projection matrix to distinguish categories; Based on the user preference type score features and the weighted features of movie popularity scores, the first y2 generalized feature vectors are selected from the feature vectors of the projection matrix; where y2 is a natural number. The filled-in movie-watching interaction features, filled-in user features, and filled-in movie features are projected onto the generalized feature vector to obtain the projected feature representation; The projection feature representation is evaluated and optimized to obtain optimized projection viewing features; The resources to be recommended in the product recommendation resource pool are updated based on the product tag information that matches the resources to be recommended and the optimized projection viewing features.

3. The product recommendation system based on moviegoer characteristics according to claim 2, characterized in that, The movie-watching feature extraction module extracts user features and movie features from the collected real-time movie-watching data of users, and is configured as follows: An autoencoder is used to encode the user's real-time viewing data to map the user's real-time viewing data into an embedding space, thereby obtaining the user's embedding representation; wherein, the number of features contained in the user's embedding representation is less than the number of features in the user's real-time viewing data. The user characteristics are determined based on the user's embedded representation; By utilizing a multimodal video feature extraction framework, sentiment analysis and topic mining are performed on users' real-time viewing data to construct a film knowledge graph; The film knowledge graph is analyzed to obtain the film features; wherein, the film features include the film's creation and performance features, visual appearance features, optical flow features and audio features; the creation and performance features include the features of the creator, the features of the performer, and the content classification features.

4. The product recommendation system based on moviegoer characteristics according to claim 2, characterized in that, The movie-watching feature extraction module extracts movie-watching interaction features from the collected real-time movie-watching data of users, and is configured as follows: Recurrent neural networks are used to model the viewing behavior sequence data in the user's real-time viewing data. Combined with scene perception algorithms, the user's viewing device, time, and location are incorporated into the viewing behavior sequence data. Based on the movie-watching behavior sequence data, determine the association between the movie-watching behavior set and the movie-watching behavior; Each movie-watching behavior in the set of movie-watching behaviors is treated as a node, and the edges between nodes represent the association weights between the movie-watching behaviors, so as to construct a user movie-watching behavior graph. The user movie-watching behavior map is analyzed by graph neural network to uncover potential patterns and rules of user movie-watching behavior and obtain movie-watching behavior characteristics; We perform text sentiment analysis on users' real-time viewing data and rating feedback data to extract users' text sentiment features and voice sentiment features when rating. The text sentiment features and the voice sentiment features are fused to obtain the rating and feedback features; By combining online learning algorithms with time series analysis algorithms, user preference features are extracted from real-time viewing data, and a user interest evolution graph is constructed. The interest evolution graph is used to record the user's current preferences and interest points, as well as to trace the source and development path of interests. Based on the interest evolution graph, the interest associations of users in different fields are mined to obtain the preferences and interest features; Construct a dynamic user-viewing social network; wherein, the user-viewing social network is used to represent users' social relationships and interactive behaviors; Based on the aforementioned user movie-watching social network, and combined with the influence propagation model, we analyze the user's influence, emotional tendencies, and information propagation paths within the social network. The social interaction characteristics are determined based on the user's influence, emotional inclination, and information dissemination path on the social network.

5. The product recommendation system based on moviegoer characteristics according to claim 2, characterized in that, The evaluation and optimization of the projection feature representation to obtain optimized projection viewing features are configured as follows: An F-test is performed on the projected feature representation to process features that do not conform to a normal distribution, and features with significant F-statistics are reconstructed. Using gradient boosting tree as the base model, recursive feature elimination is performed on the projected features after F test, eliminating the 15% of features with the smallest weight coefficient in each round. The projected features after recursive feature elimination were trained using the XGBoost model and the LightGBM model. The first importance of the projected features after recursive feature elimination in the XGBoost model and the second importance in the LightGBM model were calculated respectively. The first and second importance of the projected features after recursive feature elimination are weighted and fused to obtain the reference importance of the projected features after recursive feature elimination. Retain the top 80% of features by reference importance to form a feature candidate set; Based on the data distribution of the feature candidate set, adaptively select 7-fold cross-validation or 5-fold cross-validation; The feature candidate set is evaluated using 7-fold cross-validation or 5-fold cross-validation combined with hierarchical sampling; Based on the evaluation results of the feature candidate set, an adaptive sensitivity analysis is performed on each feature in the feature candidate set using a feature contribution analysis algorithm based on Shapley value and LIME algorithm, and a sensitivity threshold is dynamically set according to the importance and relevance of the feature. For features below the sensitivity threshold, feature pruning is performed; For the pruned feature candidate set, an adaptive feature fusion algorithm based on deep learning attention mechanism is used to fuse the features in the pruned feature candidate set to obtain the optimized projection viewing features.

6. The product recommendation system based on moviegoer characteristics according to claim 2, characterized in that, The intra-class scatter matrix is: The inter-class scatter matrix is: The projection matrix is The calculation process of the projection matrix is ​​as follows: calculate inverse matrix ; Will and Multiply, we get ; Wherein, the inverse matrix It is a square matrix; The feature vectors are used to maximize inter-class separation and minimize intra-class separation.

7. The product recommendation system based on moviegoer characteristics according to claim 2, characterized in that, The movie viewing data collection module collects real-time movie viewing data from users on different movie viewing platforms and is configured as follows: Collect data from cinema ticketing systems, online video platform viewing records, and movie-related discussions on social media. The collected data from cinema ticketing systems, online video platform viewing records, and movie-related discussions on social media are integrated into real-time movie-watching data for users.

Images