Artificial intelligence intelligent recommendation system and method

Abstract

The invention relates to the technical field of intelligent recommendation, in particular to an artificial intelligence intelligent recommendation system and method.In the invention, a data acquisition module acquires a user behavior log including multi-modal behavior data and article feature data from a shared account; the behavior pattern recognition module is used for recognizing unique behavior patterns of different users by using a clustering algorithm and generating an initial behavior feature vector of each user; the time sequence analysis module combines the feature vectors with time factors to generate a time sequence preference matrix and time sequence behavior feature vectors, and generates a user preference matrix through a matrix decomposition method; the article time sequence feature matrix generation unit generates an article time sequence feature matrix according to the article feature data and the time factors; and a prediction model generation module fuses the user time sequence preference matrix and the article time sequence feature matrix to generate a multilayer perceptron prediction model, and by optimizing hyper-parameters and minimizing prediction errors, the personalized recommendation effect in a multi-user shared account scene is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent recommendation technology, and in particular to an artificial intelligence intelligent recommendation system and method. Background Art

[0002] Traditional intelligent recommendation systems face many technical problems when dealing with multiple users sharing accounts.

[0003] First, traditional systems usually assume that each account corresponds to an independent user. Therefore, different users under the same account cannot be effectively distinguished during the data acquisition and processing stages, resulting in multimodal behavior data in behavior logs (such as ratings, browsing time, number of clicks, search history, and viewing time, etc.) being mixed together, losing the basis for personalized recommendations.

[0004] Secondly, due to the confusion of user behavior data, traditional systems cannot accurately capture the behavior patterns of each user when generating user behavior feature vectors, resulting in the generated feature vectors being unable to effectively represent the behavior habits of specific users, thus affecting the accuracy of recommendation results. In addition, traditional systems also have limitations in time series analysis and are unable to effectively combine time factors with user behavior patterns, resulting in a lack of capture of dynamic changes in user behavior in recommendation results.

[0005] Finally, due to the lack of a dedicated processing mechanism for shared account environments, traditional recommendation systems tend to use global models or methods based on static features when generating prediction models, which cannot accurately predict the behavioral preferences of different users at different time segments, further weakening the personalization and accuracy of recommendations. These problems are particularly prominent in scenarios where multiple users share accounts, seriously affecting user experience and system effectiveness. Summary of the invention

[0006] The purpose of the present invention is to provide an artificial intelligence intelligent recommendation system and method to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, one of the objectives of the present invention is to provide an artificial intelligence intelligent recommendation system, including a data acquisition module, a behavior pattern recognition module, a time series analysis module and a prediction model generation module, wherein: The data acquisition module acquires a user behavior log from a shared account, wherein the user behavior log includes multimodal behavior data and item feature data; The behavior pattern recognition module uses a clustering algorithm to recognize unique behavior patterns of different users in the multimodal behavior data and generates initial behavior feature vectors of different users; The time series analysis module combines the unique behavior patterns of different users in the multimodal behavior data with the time factor to generate the time series preference matrix of different users; performs time series analysis on the initial behavior feature vectors of different users to generate the time series behavior feature vectors of different users; and uses the matrix decomposition method to map the time series behavior feature vectors of different users to the time series preference matrix to generate the user preference matrix; constructs the item feature matrix according to the item feature data, and combines the time factor with the matrix to generate the item time series feature matrix; The prediction model generation module generates a prediction model according to the temporal preference matrix of different users and the temporal feature matrix of items.

[0008] As a further improvement of the technical solution, the process of generating initial behavior feature vectors of different users by the behavior pattern recognition module specifically includes: Extracting multimodal behavior data from the user behavior log provided by the data acquisition module, wherein the multimodal behavior data includes multiple behavior feature vectors; Determine the number of clusters and select the optimal K value using the elbow method; K refers to the number of clusters, indicating the number of user behavior pattern categories that need to be divided.

[0009] Initialize K cluster centers and select the initial cluster centers randomly; Iteratively assign each user's behavior feature vector to the class to which the nearest cluster center belongs; Recalculate the cluster center of each class, that is, the average value of the behavioral feature vectors of all users in the class; Repeat the above steps of assigning and recalculating cluster centers until the cluster centers no longer change; After each user's behavior feature vector is assigned to a cluster center, the user's initial behavior feature vector is the cluster center of the class to which it belongs.

[0010] As a further improvement of the technical solution, the timing analysis module includes a timing preference matrix generation unit, and the process of the timing preference matrix generation unit generating timing preference matrices of different users specifically includes: Extract multimodal behavior data and timestamp information from the user behavior log provided by the data acquisition module; segment the timestamp information at fixed time intervals to generate time segment labels; construct a temporal preference matrix based on user and time segment labels; the rows of the matrix represent users, the columns represent time segments, and the values ​​in the cells represent the average ratings of the user's behavior feature vectors in the time segment.

[0011] As a further improvement of the present technical solution, the time series analysis module includes a user preference matrix generation unit, and the process of the user preference matrix generation unit generating the user preference matrix specifically includes: Extract the behavior feature vectors of each user in different time segments from the generated time series preference matrix; use the time series analysis method to analyze the user's behavior feature vectors, extract the periodicity, trend and randomness features in the time series, and use them together as the time series behavior feature vectors; Generate time series behavior feature vectors of different users, which can capture the dynamic changes of user behavior; the time series behavior feature vector includes the summary value of the user's behavior feature vectors in different time segments and the periodicity, trend and randomness characteristics in the time series; The generated time series behavior feature vector is combined with the time series preference matrix, and the time series behavior feature vector is mapped to the time series preference matrix using the matrix decomposition method to generate the user preference matrix; the dimension and parameters of the decomposition matrix are initialized; the time series preference matrix is ​​decomposed to generate the user feature matrix and the time feature matrix; the generated time series behavior feature vector is combined with the user feature matrix to generate the user preference matrix; the decomposition parameters are optimized by minimizing the error between the predicted value and the actual value.

[0012] As a further improvement of the technical solution, the time series analysis module includes an item time series feature matrix generation unit, and the process of the item time series feature matrix generation unit generating the item time series feature matrix specifically includes: Extracting time-related features from the item feature data provided by the data acquisition module; Based on the item feature data and time segment labels, an item time series feature matrix is ​​constructed; the rows of the matrix represent items, the columns represent time segments, and the values ​​in the cells represent the feature values ​​of the items in the time segment; By taking the time segment labels as the columns of the matrix and the characteristic values ​​of the items in each time segment as the values ​​in the cells, the time factor is combined with the item feature matrix to generate the item time series feature matrix.

[0013] As a further improvement of the present technical solution, the process of generating a prediction model by the prediction model generation module specifically includes: The user behavior feature vectors in the temporal preference matrix of different users and the item feature vectors in the item temporal feature matrix are merged to generate a user-item combination feature vector; Select multi-layer perceptron as the prediction model; divide the user-item combination feature vector and known rating data into training set and test set; The hyperparameters of the multilayer perceptron are optimized using the grid search method; the best hyperparameter combination is selected through cross-validation; the prediction model is trained using the training set, and the model parameters are optimized by minimizing the error between the predicted score and the actual score; the mean square error is selected as the loss function, and the model parameters are adjusted through the back-propagation algorithm; During the training process, the model parameters are continuously adjusted until the prediction error of the prediction model is minimized; The trained prediction model parameters are saved to generate the final prediction model; the final prediction model outputs the predicted score of each user for each item in different time segments based on the user time series preference matrix and the item time series feature matrix.

[0014] The second object of the present invention is to provide a method for an artificial intelligence intelligent recommendation system, comprising the following method steps: S1. Obtaining a user behavior log from a shared account, wherein the user behavior log includes multimodal behavior data and item feature data; S2, using clustering algorithms to identify unique behavior patterns of different users in multimodal behavior data, and generate initial behavior feature vectors of different users; S3. Combine the unique behavior patterns of different users in the multimodal behavior data with the time factor to generate the time series preference matrix of different users; perform time series analysis on the initial behavior feature vectors of different users to generate the time series behavior feature vectors of different users; and use the matrix decomposition method to map the time series behavior feature vectors of different users to the time series preference matrix to generate the user preference matrix; construct the item feature matrix based on the item feature data, and combine the time factor with the matrix to generate the item time series feature matrix; S4. Generate a prediction model based on the temporal preference matrix of different users and the temporal feature matrix of items.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The artificial intelligence intelligent recommendation system and method generates an initial behavior feature vector by extracting multimodal behavior data from shared accounts and using clustering algorithms to identify the unique behavior patterns of different users; this process effectively solves the primary challenge of personalized recommendation when multiple users share accounts, that is, how to distinguish and identify the behavior patterns of different users to ensure the personalization and accuracy of the recommendation results.

[0016] 2. The artificial intelligence intelligent recommendation system and method combines the user's behavioral feature vector with the time factor to generate a temporal preference matrix and an item temporal feature matrix, and uses advanced matrix decomposition and multi-layer perceptron models for prediction; this method not only captures the dynamic changes in user behavior, but also takes into account the changes in item characteristics over time, significantly improving the accuracy and stability of the recommendation model, and providing users with recommendation results that are more in line with their current needs and interests, thereby enhancing user experience and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the overall method flow of the present invention; Figure 2 It is a schematic diagram of the overall module unit of the present invention.

[0018] In the figure: 100, data acquisition module; 200, behavior pattern recognition module; 300, time series analysis module; 301, time series preference matrix generation unit; 302, user preference matrix generation unit; 303, item time series feature matrix generation unit; 400, prediction model generation module. DETAILED DESCRIPTION

[0019] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Next, see Figure 1-Figure 2 One of the purposes of this embodiment is to provide an artificial intelligence intelligent recommendation system, including a data acquisition module 100, a behavior pattern recognition module 200, a time series analysis module 300 and a prediction model generation module 400.

[0021] The data acquisition module 100 acquires the user behavior log from the shared account, wherein the user behavior log includes multimodal behavior data and item feature data, specifically including: Connect to a shared account and obtain user behavior logs through API or database query; format the obtained user behavior logs and item feature data into a processable structure; and pre-process multimodal behavior data and item feature data, such as removing invalid records and processing missing values.

[0022] The behavior pattern recognition module 200 uses a clustering method to identify the unique behavior patterns of different users in the multimodal behavior data and generates initial behavior feature vectors of different users, specifically including: Extracting multimodal behavior data from the user behavior log provided by the data acquisition module 100, the multimodal behavior data including multiple behavior feature vectors such as user ratings, browsing time, number of clicks, search records, and viewing time; Determine the number of clusters and select the optimal K value through the elbow method. K refers to the number of clusters, which indicates the number of user behavior pattern categories that need to be divided. Specifically, the K value determines how many different classes the user's behavior feature vector is divided into, and each class represents a unique behavior pattern. Initialize K cluster centers and select the initial cluster centers randomly; Iteratively assign each user's behavior feature vector to the class to which the nearest cluster center belongs; Recalculate the cluster center of each class, that is, the average value of the behavioral feature vectors of all users in the class; Repeat the above steps of assigning and recalculating cluster centers until the cluster centers no longer change; After each user's behavior feature vector is assigned to a cluster center, the user's initial behavior feature vector is the cluster center of the class to which it belongs; the generated initial behavior feature vector can represent the user's unique behavior pattern in the multimodal behavior data, providing basic data for subsequent time series analysis and personalized recommendations.

[0023] The behavior pattern recognition module 200 uses a clustering method to identify the unique behavior patterns of different users from multimodal behavior data and generates an initial behavior feature vector for each user; the optimal K value is determined by the elbow method and multiple iterations are performed. The final generated initial behavior feature vector can accurately represent the behavior habits of each user, solving the problem that traditional systems cannot effectively capture user behavior patterns due to data congestion, and providing basic data for subsequent time series analysis and personalized recommendations.

[0024] The temporal preference matrix generation unit 301 in the temporal analysis module 300 combines the unique behavior patterns of different users in the multimodal behavior data with the time factor to generate the temporal preference matrix of different users, which specifically includes: Extract multimodal behavior data and timestamp information from the user behavior log provided by the data acquisition module 100; segment the timestamp information according to fixed time intervals (such as hours, days, weeks, etc.) to generate time segment labels; construct a temporal preference matrix based on the user and time segment labels; the rows of the matrix represent users, the columns represent time segments, and the values ​​in the cells represent the average scores of the user's behavior feature vectors in the time segment.

[0025] The temporal preference matrix generation unit 301 extracts the multimodal behavior data and timestamp information in the user behavior log, generates time segment labels at fixed time intervals, and constructs a temporal preference matrix; the rows of the matrix represent users, the columns represent time segments, and the values ​​in the cells are the average scores of the user's behavior feature vectors in the time segment; this unit effectively solves the limitations of traditional systems in time series analysis, ensures that the recommendation system can capture the dynamic changes in user behavior, and improves the timeliness and accuracy of recommendation results.

[0026] The user preference matrix generation unit 302 in the time series analysis module 300 performs time series analysis on the initial behavior feature vectors of different users to generate time series behavior feature vectors of different users; and uses a matrix decomposition method to map the time series behavior feature vectors of different users to the time series preference matrix to generate a user preference matrix, specifically including: Extract the behavior feature vectors of each user in different time segments from the generated time series preference matrix; use the time series analysis method to analyze the user's behavior feature vectors, extract the periodicity, trend and randomness features in the time series, and use them together as the time series behavior feature vectors; Generate time series behavior feature vectors of different users, which can capture the dynamic changes of user behavior; the time series behavior feature vector includes the summary value of the user's behavior feature vectors in different time segments and the periodicity, trend and randomness characteristics in the time series; The generated time series behavior feature vector is combined with the time series preference matrix, and the time series behavior feature vector is mapped to the time series preference matrix using the matrix decomposition method to generate a user preference matrix; the dimension and parameters of the decomposition matrix are initialized; the time series preference matrix is ​​decomposed to generate a user feature matrix and a time feature matrix; the generated time series behavior feature vector is combined with the user feature matrix to generate a user preference matrix; the decomposition parameters are optimized by minimizing the error between the predicted value and the actual value to ensure that the generated user preference matrix can accurately reflect the user's comprehensive preferences.

[0027] The user preference matrix generation unit 302 performs time series analysis on the behavior feature vectors in the time series preference matrix, extracts periodicity, trend and randomness features, generates time series behavior feature vectors, and uses matrix decomposition method to map these feature vectors to the time series preference matrix to generate a user preference matrix; this unit not only captures the dynamic changes of user behavior, but also optimizes the representation of feature vectors through matrix decomposition, solving the problem that traditional systems cannot consider time and dynamic features when generating user preference matrices, and further improves the personalization and accuracy of recommendations.

[0028] The item time series feature matrix generation unit 303 in the time series analysis module 300 constructs an item feature matrix according to the item feature data, and combines the time factor with the matrix to generate an item time series feature matrix, which specifically includes: Extracting time-related features from the item feature data provided by the data acquisition module 100, such as the release date and update frequency of the item; Based on the item feature data and time segment labels, an item time series feature matrix is ​​constructed; the rows of the matrix represent items, the columns represent time segments, and the values ​​in the cells represent the feature values ​​of the item in the time segment (such as the number of releases, update frequency, etc.); By taking the time segment labels as the columns of the matrix and the characteristic values ​​of the items in each time segment as the values ​​in the cells, the time factor is combined with the item feature matrix to generate the item time series feature matrix.

[0029] The item time series feature matrix generation unit 303 extracts time-related features, such as the release date and update frequency of the item, from the item feature data provided by the data acquisition module 100, and generates time segment labels at fixed time intervals to construct an item time series feature matrix; the rows of the matrix represent items, the columns represent time segments, and the values ​​in the cells are the feature values ​​of the item in the time segment; this unit effectively solves the problem of the lack of time factors in traditional systems when processing item features, ensures that the recommendation system can capture the dynamic changes of item features, and improves the accuracy and relevance of recommendations.

[0030] The prediction model generation module 400 generates a prediction model according to the temporal preference matrix of different users and the temporal feature matrix of items, which specifically includes: The user behavior feature vectors in the temporal preference matrix of different users and the item feature vectors in the item temporal feature matrix are merged to generate a user-item combination feature vector; Multilayer perceptron is selected as the prediction model. Multilayer perceptron can handle complex nonlinear relationships and is suitable for prediction tasks of multimodal data. Divide the user-item combination feature vector and known rating data into a training set and a test set; usually 70% of the data is used as the training set and 30% of the data is used as the test set; Use grid search or random search to optimize the hyperparameters of the neural network; the specific hyperparameters optimized include learning rate, number of hidden layers, number of neurons in each layer, activation function, etc. Select the best hyperparameter combination through cross-validation to ensure the stable performance of the model on different data subsets; Use the training set to train the prediction model and optimize the model parameters by minimizing the error between the predicted score and the actual score. The specific goal is to make the output of the prediction model as close as possible to the actual user rating data. Select the mean square error as the loss function and adjust the model parameters through the back propagation algorithm. During the training process, the model parameters are continuously adjusted until the prediction error of the prediction model reaches the minimum or the preset training rounds are met; The trained prediction model parameters are saved to generate the final prediction model; the final prediction model can output each user's predicted score for each item in different time segments based on the user's temporal preference matrix and the item's temporal feature matrix, thereby providing accurate data support for the personalized recommendation system.

[0031] The prediction model generation module 400 generates a user-item combination feature vector by fusing the temporal preference matrices of different users and the temporal feature matrices of items, and selects a multilayer perceptron as the prediction model; it optimizes hyperparameters through grid search and cross-validation, minimizes the error between the predicted score and the actual score, and generates the final prediction model; this module not only handles the complex nonlinear relationship of multimodal data, but also ensures the high accuracy and stability of the model by optimizing parameters, solving the problem that traditional systems tend to use global or static methods when generating prediction models, and significantly improving the effect of personalized recommendations and user experience.

[0032] The second purpose of this embodiment is to provide a method of an artificial intelligence intelligent recommendation system, including the following method steps: S1. Obtaining a user behavior log from a shared account, wherein the user behavior log includes multimodal behavior data and item feature data; S2, using clustering algorithms to identify unique behavior patterns of different users in multimodal behavior data, and generate initial behavior feature vectors of different users; S3. Combine the unique behavior patterns of different users in the multimodal behavior data with the time factor to generate the time series preference matrix of different users; perform time series analysis on the initial behavior feature vectors of different users to generate the time series behavior feature vectors of different users; and use the matrix decomposition method to map the time series behavior feature vectors of different users to the time series preference matrix to generate the user preference matrix; construct the item feature matrix based on the item feature data, and combine the time factor with the matrix to generate the item time series feature matrix; S4. Generate a prediction model based on the temporal preference matrix of different users and the temporal feature matrix of items.

[0033] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. Artificial intelligence intelligent recommendation system, characterized by: The system comprises a data acquisition module (100), a behavior pattern recognition module (200), a time series analysis module (300) and a prediction model generation module (400), wherein: The data acquisition module (100) acquires a user behavior log from a shared account, wherein the user behavior log includes multimodal behavior data and item feature data; The behavior pattern recognition module (200) uses a clustering algorithm to recognize unique behavior patterns of different users in the multimodal behavior data, and generates initial behavior feature vectors of different users; The time series analysis module (300) combines the unique behavior patterns of different users in the multimodal behavior data with the time factor to generate a time series preference matrix for different users; performs time series analysis on the initial behavior feature vectors of different users to generate time series behavior feature vectors for different users; and uses a matrix decomposition method to map the time series behavior feature vectors of different users onto the time series preference matrix to generate a user preference matrix; constructs an item feature matrix based on the item feature data, and combines the time factor with the matrix to generate an item time series feature matrix; The prediction model generation module (400) generates a prediction model according to the temporal preference matrix of different users and the temporal feature matrix of items.

2. The artificial intelligence intelligent recommendation system according to claim 1, characterized in that: The process of the behavior pattern recognition module (200) generating initial behavior feature vectors of different users specifically includes: Extracting multimodal behavior data from the user behavior log provided by the data acquisition module (100), wherein the multimodal behavior data includes multiple behavior feature vectors; Determine the number of clusters and select the optimal K value using the elbow method. K refers to the number of clusters, which indicates the number of user behavior pattern categories that need to be divided. Initialize K cluster centers and select the initial cluster centers randomly; Iteratively assign each user's behavior feature vector to the class to which the nearest cluster center belongs; Recalculate the cluster center of each class, that is, the average value of the behavioral feature vectors of all users in the class; Repeat the above steps of assigning and recalculating cluster centers until the cluster centers no longer change; After each user's behavior feature vector is assigned to a cluster center, the user's initial behavior feature vector is the cluster center of the class to which it belongs.

3. The artificial intelligence intelligent recommendation system according to claim 1, characterized in that: The time series analysis module (300) comprises a time series preference matrix generation unit (301), and the process of the time series preference matrix generation unit (301) generating time series preference matrices of different users specifically comprises: Extracting multimodal behavior data and timestamp information from the user behavior log provided by the data acquisition module (100); segmenting the timestamp information at fixed time intervals to generate time segment labels; constructing a temporal preference matrix based on the user and time segment labels; the rows of the matrix represent users, the columns represent time segments, and the values ​​in the cells represent the average ratings of the user's behavior feature vectors in the time segment.

4. The artificial intelligence intelligent recommendation system according to claim 3, characterized in that: The time series analysis module (300) comprises a user preference matrix generation unit (302), and the process of generating the user preference matrix by the user preference matrix generation unit (302) specifically comprises: Extract the behavior feature vectors of each user in different time segments from the generated time series preference matrix; use the time series analysis method to analyze the user's behavior feature vectors, extract the periodicity, trend and randomness features in the time series, and use them together as the time series behavior feature vectors; Generate time series behavior feature vectors of different users, which can capture the dynamic changes of user behavior; the time series behavior feature vector includes the summary value of the user's behavior feature vectors in different time segments and the periodicity, trend and randomness characteristics in the time series; The generated time series behavior feature vector is combined with the time series preference matrix, and the time series behavior feature vector is mapped to the time series preference matrix using the matrix decomposition method to generate the user preference matrix; the dimension and parameters of the decomposition matrix are initialized; the time series preference matrix is ​​decomposed to generate the user feature matrix and the time feature matrix; the generated time series behavior feature vector is combined with the user feature matrix to generate the user preference matrix; the decomposition parameters are optimized by minimizing the error between the predicted value and the actual value.

5. The artificial intelligence intelligent recommendation system according to claim 4, characterized in that: The time series analysis module (300) comprises an item time series feature matrix generation unit (303), and the process of generating the item time series feature matrix by the item time series feature matrix generation unit (303) specifically comprises: Extracting time-related features from the item feature data provided by the data acquisition module (100); Based on the item feature data and time segment labels, an item time series feature matrix is ​​constructed; the rows of the matrix represent items, the columns represent time segments, and the values ​​in the cells represent the feature values ​​of the items in the time segment; By taking the time segment labels as the columns of the matrix and the characteristic values ​​of the items in each time segment as the values ​​in the cells, the time factor is combined with the item feature matrix to generate the item time series feature matrix.

6. The artificial intelligence intelligent recommendation system according to claim 5, characterized in that: The process of generating a prediction model by the prediction model generation module (400) specifically includes: The user behavior feature vectors in the temporal preference matrix of different users and the item feature vectors in the item temporal feature matrix are merged to generate a user-item combination feature vector; Select multi-layer perceptron as the prediction model; divide the user-item combination feature vector and known rating data into training set and test set; The hyperparameters of the multilayer perceptron are optimized using the grid search method; the best hyperparameter combination is selected through cross-validation; the prediction model is trained using the training set, and the model parameters are optimized by minimizing the error between the predicted score and the actual score; the mean square error is selected as the loss function, and the model parameters are adjusted through the back-propagation algorithm; During the training process, the model parameters are continuously adjusted until the prediction error of the prediction model is minimized; The trained prediction model parameters are saved to generate the final prediction model; the final prediction model outputs the predicted score of each user for each item in different time segments based on the user time series preference matrix and the item time series feature matrix.

7. A method for using the artificial intelligence intelligent recommendation system according to any one of claims 1 to 6, characterized in that: The method comprises the following steps: S1. Obtaining a user behavior log from a shared account, wherein the user behavior log includes multimodal behavior data and item feature data; S2, using clustering algorithms to identify unique behavior patterns of different users in multimodal behavior data, and generate initial behavior feature vectors of different users; S3. Combine the unique behavior patterns of different users in the multimodal behavior data with the time factor to generate the time series preference matrix of different users; perform time series analysis on the initial behavior feature vectors of different users to generate the time series behavior feature vectors of different users; and use the matrix decomposition method to map the time series behavior feature vectors of different users to the time series preference matrix to generate the user preference matrix; construct the item feature matrix based on the item feature data, and combine the time factor with the matrix to generate the item time series feature matrix; S4. Generate a prediction model based on the temporal preference matrix of different users and the temporal feature matrix of items.

Images