Regional differentiation recommendation system for convergence media content
By integrating multiple recommendation models and geographic information systems, combining user geographical location and content regional characteristics, the problem of insufficient regional differentiation in the integrated media content recommendation system is solved, precise and personalized content recommendation is achieved, and user experience and information dissemination efficiency is improved.
Patent Information
- Application Number
- CN202510590515.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-08
- Publication Date
- 2025-08-15
AI Technical Summary
When handling regional differentiated recommendations, the existing integrated media content recommendation system ignores the regional characteristics of the content and the user's geographical location information, resulting in the lack of geographical targeting and timeliness of recommended content, which cannot meet the user's personalized information needs in different regions. The modeling process of user interest preferences is complex and susceptible to noise data, resulting in limited recommendation accuracy.
A variety of recommendation models are adopted, including collaborative filtering, content filtering and deep learning models, combining the user's geographical location, interest preferences and behavior history, and obtain user latitude and longitude coordinates through the geographical information system, build user portraits, and combine the regional characteristics and timeliness of the content to carry out intelligent regional differentiated recommendations.
Accurate and personalized content recommendations have been achieved, improving user experience and information dissemination efficiency, ensuring that the recommended content meets users' personalized needs and has distinct regional characteristics.
Smart Images
Figure CN120492897A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of converged media technology, and specifically relates to a regional differentiated recommendation system for converged media content. Background Art
[0002] Converged media refers to media content that integrates multiple media formats (such as text, images, audio, and video) to provide a comprehensive, multi-faceted approach to information dissemination. With the rapid development of the internet, information overload has become increasingly severe, making it increasingly difficult for users to obtain effective information. To address this issue, various content recommendation systems have emerged, such as those based on collaborative filtering, content filtering, and deep learning. These systems extract and analyze features from various forms of converged media, improving the accuracy and efficiency of content recommendations to a certain extent. However, they still have many shortcomings.
[0003] Currently, mainstream content recommendation systems on the market primarily employ multi-dimensional recommendation strategies based on user behavior data, content features, and social network information. These systems use machine learning algorithms to model user interests and preferences and make content recommendations based on the model's predictions. However, when making regionally differentiated recommendations, these systems often ignore the regional characteristics of content and the user's location information, resulting in recommendations that lack regional relevance and timeliness. Furthermore, they lack an effective regionally differentiated recommendation mechanism and are unable to meet the personalized information needs of users in different regions. Furthermore, insufficient consideration of the regional characteristics and timeliness of content leads to a mismatch between recommended content and actual demand. Furthermore, the process of modeling user interests and preferences is complex and susceptible to noisy data, resulting in limited recommendation accuracy. Summary of the Invention
[0004] In response to the above-mentioned shortcomings, the present invention discloses a regional differentiated recommendation system for converged media content. By integrating and optimizing multiple recommendation models, it can achieve more accurate and personalized content recommendations than the existing technology. It can make intelligent regional differentiated recommendations based on multi-dimensional information such as the user's geographic location, interest preferences, behavioral history, as well as the regional characteristics, timeliness, popularity and other attributes of the content, to meet the personalized information needs of users in different regions and at different times, and improve user experience and information dissemination efficiency.
[0005] The present invention is achieved by adopting the following technical solutions:
[0006] A regional differentiated recommendation system for converged media content, comprising a central processing unit, a data collection module, a data vectorization module, a user feature extraction module, a content feature extraction module, a recommendation model integration module, a regional differentiated recommendation module, and a data storage module;
[0007] The central processing unit is used to coordinate the work of various modules;
[0008] The data collection module is used to collect user behavior data;
[0009] The data vectorization module is used to convert data into vector form;
[0010] The user feature extraction module includes a geographic location extraction module and an interest preference module. The geographic location extraction module uses GIS technology to obtain the user's latitude and longitude coordinates (x, y) and stores them in the data storage module. The interest preference module extracts the user's interest preferences and behavior pattern feature data based on the user's behavior data and stores the feature data in the data storage module after processing by the data vectorization module.
[0011] The content feature extraction module includes a semantic feature extraction module, an image content feature extraction module, an audio feature extraction module, a time and region extraction module, and a content portrait construction module. The semantic feature extraction module uses a semantic analysis algorithm to process the text content to extract semantic feature data and stores the data in the data storage module after processing by the data vectorization module. The image content feature extraction module uses an image recognition model in the field of computer vision (CV, Computer Vision) to extract features from the picture content, recognizes the feature data of objects, scenes, and colors in the image, and stores the data in the data storage module after processing by the data vectorization module. The audio feature extraction module uses an audio feature extraction algorithm in audio processing technology (Audio Feature Extraction Technology). Extraction Algorithm) identifies the characteristic data of the tone, rhythm, and emotion of the audio content and stores them in the data storage module after processing by the data vectorization module; the time and region extraction module extracts the characteristic data of the time information and region information of the content release and stores them in the data storage module after processing by the data vectorization module; the content profile construction module collects the characteristic data obtained in the semantic feature extraction module, the image content feature extraction module, the audio feature extraction module, and the time and region extraction module to obtain the content profile (Content Profiling) of the corresponding content;
[0012] The recommendation model integration module is constructed based on a combination of one or more models selected from the group consisting of a collaborative filtering model, a content-based filtering model, and a deep learning model, and selects corresponding models to process and analyze the feature data of the user feature extraction module and the content extraction module to obtain feature data of content that meets the user's interest preferences;
[0013] The data storage module includes a working data storage module and a knowledge base. The working data storage module is used to store data obtained from each module, and the knowledge base is used to store the vectorized converged media content.
[0014] The regional differentiated recommendation module determines the region to which the user belongs based on the user's latitude and longitude coordinates (x, y), selects the regional feature vector of the corresponding region, and obtains the feature data of the recommended content based on the feature data of the content that meets the user's interest preferences obtained by the recommendation model integration module, and then retrieves the corresponding content from the knowledge base based on the feature data of the recommended content for push.
[0015] Furthermore, the user's behavior data includes the user's clicks, browsing, purchases, searches, historical browsing records, interactive behaviors, and basic information of the user.
[0016] Furthermore, in the interest preference module, it is assumed that user U's interest preference for category C is P u，c , the calculation formula is as follows: u，c =Number of user U's behaviors related to category C / total number of user U's behaviors.
[0017] Furthermore, the semantic feature data includes semantic feature data of keywords, topics and sentiment tendencies; the semantic feature extraction module adopts the following formula:
[0018] Formula I: where n i,j is the number of times word i appears in document j, ∑ k n i,j is the total number of words in document j, TF i,j is the word frequency;
[0019] Formula II: Where N is the total number of documents, n i is the number of documents containing word i, IDF i is the inverse document frequency;
[0020] Formula III (Text Content Feature Extraction): TF-IDFi,j =TF i,j ×IDF i .
[0021] Furthermore, in the recommendation model integration module, a collaborative filtering model is adopted based on the user-item rating matrix (User-ItemRatingMatrix), and a similarity calculation function is used to measure the similarity between users or items. A content filtering model is adopted based on the feature data obtained in the content feature extraction module, and a classification algorithm or regression function is used to analyze the user's preference for the content. A deep learning model such as a convolutional neural network, a recurrent neural network and its variants is adopted to construct a deep neural network architecture (Deep Neural Network Architecture) to automatically learn the complex feature representation of user feature data and content feature data; and in the model fusion process, the recommendation model integration module adopts an ensemble learning algorithm (Ensemble Learning Algorithm) to perform weighted fusion on the prediction results of multiple models. The ensemble learning algorithm includes stacking generalization (Stacking), bagging (Bagging), and boosting (Boosting).
[0022] Furthermore, the regional feature vector is obtained by vectorizing the data of the geographical location, cultural background, and language habits in the corresponding region.
[0023] The method for using the above-mentioned regional differentiated recommendation system for integrated media content includes the following steps:
[0024] S1. Collecting converged media content data, removing duplication, removing noise, unifying the format, filling in missing data, and standardizing or normalizing the data to form a knowledge base, and processing the converged media content through the content feature extraction module to obtain feature data, and vectorizing the feature data and storing it in the knowledge base accordingly;
[0025] S2. With the user's authorization, collect the user's behavior data through a logging system, user tracking technology, and online questionnaires, and process them through the user feature extraction module to obtain the user's latitude and longitude coordinates (x, y) and feature data of the user's interests, preferences, and behavior patterns;
[0026] S3, processing and analyzing the characteristic data of the user's interest preferences and behavior patterns obtained in step S2 by the recommendation model integration module to obtain characteristic data of content that meets the user's interest preferences;
[0027] S4. Determine the region to which the user belongs based on the user's latitude and longitude coordinates (x, y) through the regional differentiation recommendation module, select the regional feature vector of the corresponding regional area, and obtain the feature data of the recommended content based on the feature data of the content that meets the user's interest preferences obtained by the recommendation model integration module, and then retrieve the corresponding content from the knowledge base based on the feature data of the recommended content for push.
[0028] Furthermore, in step S4, the retrieved integrated media content is displayed to the user in the form of a recommendation column, pop-up window, or personalized homepage, and a user feedback mechanism is provided to collect user satisfaction with the recommendation results and improvement suggestions.
[0029] Compared with the existing technology, this technical solution has the following beneficial effects:
[0030] 1. The present invention sets up a user feature extraction module, in which the geographic location extraction module accurately locates the user's geographic location, and the interest preference module extracts feature vectors that can accurately characterize the user's interest preferences and behavior patterns from the user's historical browsing records, interactive behaviors, social relationships and other multi-dimensional data. Then, the user's geographic location information and the feature vector data of the user's interest preferences and behavior patterns are combined to obtain an accurate user profile (User Profiling), which not only covers the user's basic attributes, but also deeply reflects the user's potential needs and dynamic changes, providing support for subsequent content recommendations.
[0031] 2. The present invention sets up a content feature extraction module composed of a semantic feature extraction module, an image content feature extraction module, an audio feature extraction module, a time and region extraction module, a content portrait construction module, etc., to perform in-depth analysis and feature characterization of the integrated media content, and construct multi-dimensional feature vectors such as the regional characteristics, timeliness, and popularity of the content, providing a comprehensive and detailed content portrait (Content Profiling) for content recommendation.
[0032] 3. The present invention sets a recommendation model integration module that integrates a variety of advanced recommendation models, including collaborative filtering models, content filtering models, deep learning models, etc. By constructing a multi-model fusion framework and using the model selection strategy and weight optimization algorithm, different recommendation models are dynamically selected and integrated according to user characteristics and content characteristics to improve the accuracy and robustness of recommendations.
[0033] 4. The present invention sets up a regional differentiated recommendation module to differentiate the interest preferences of users in different regions based on the user's region or designated interest area, as well as the prediction results of the recommendation model, to ensure that the recommended content not only meets the user's personalized needs but also has distinct regional characteristics, thereby realizing regional differentiated content recommendation services. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is a flowchart of a method for using the regional differentiated recommendation system for integrated media content according to the present invention. DETAILED DESCRIPTION
[0035] The present invention is further illustrated by the following examples, which are not intended to limit the present invention. Specific experimental conditions and methods not specified in the following examples are conventional methods well known to those skilled in the art.
[0036] At the same time, the following embodiments are applied in conjunction with the three-level interconnected knowledge base developed and established by the applicant of the present invention. The three-level interconnected knowledge base integrates the technical platform resources at the Guangxi Zhuang Autonomous Region level to build a unified integrated media business platform. The platform accommodates Guangxi's media history and new manuscript resources, and integrates the distributed knowledge graph of provincial-municipal-district-county integrated media platform data through technical means. In addition, by integrating the Embedding model, applying the Rerank model to optimize the search results, using TTS technology to realize voice interaction, integrating the LLM model to enhance dialogue capabilities, and building a RAG pipeline to realize knowledge reasoning, an efficient and intelligent knowledge management and retrieval system is constructed, which has the characteristics of ease of use, performance optimization, rich functional components, support for multiple large language models, and strict security measures.
[0037] Example 1: A regional differentiated recommendation system for integrated media content, which includes a central processing unit, a data collection module, a data vectorization module, a user feature extraction module, a content feature extraction module, a recommendation model integration module, a regional differentiated recommendation module, and a data storage module; the central processing unit is used to coordinate the work of each module; the data collection module is used to collect user behavior data; the data vectorization module is used to convert data into vector form; the user feature extraction module includes a geographic location extraction module and an interest preference module, the geographic location extraction module uses GIS technology to obtain the user's latitude and longitude coordinates (x, y) and stores them in the data storage module, the interest preference module extracts the user's interest preferences and behavior pattern feature data based on the user's behavior data and stores them in the data storage module after processing by the data vectorization module; the user's behavior data includes the user's clicks, browsing, purchases, searches, the user's historical browsing records, the user's interactive behavior, and the user's basic information; in the interest preference module, it is assumed that the user U's interest preference for category C is P u，c , the calculation formula is as follows: u，c = the number of user U's actions related to category C / the total number of user U's actions;
[0038] The content feature extraction module includes a semantic feature extraction module, an image content feature extraction module, an audio feature extraction module, a time and region extraction module, and a content portrait construction module. The semantic feature extraction module uses a semantic analysis algorithm to process the text content to extract semantic feature data, and stores the extracted semantic feature data in the data storage module after processing by the data vectorization module. The semantic feature data includes semantic feature data of keywords, themes, and emotional tendencies. The semantic feature extraction module uses the following formula:
[0039] Formula I: where n i,j is the number of times word i appears in document j, ∑ k n i,j is the total number of words in document j, TF i,j is the word frequency;
[0040] Formula II: Where N is the total number of documents, n i is the number of documents containing word i, IDF i is the inverse document frequency;
[0041] Formula III (Text Content Feature Extraction): TF-IDF i,j =TF i,j ×IDF i ;
[0042] The image content feature extraction module uses the image recognition model in the field of computer vision to extract features from the picture content, identifies the feature data of objects, scenes, and colors in the image, and stores them in the data storage module after processing by the data vectorization module. The audio feature extraction module uses the audio feature extraction algorithm in the audio processing technology to identify the feature data of the tone, rhythm, and emotion of the audio content, and stores them in the data storage module after processing by the data vectorization module. The time and region extraction module extracts the feature data of the time information and region information of the content release, and stores them in the data storage module after processing by the data vectorization module. The content portrait construction module collects the feature data obtained in the semantic feature extraction module, the image content feature extraction module, the audio feature extraction module, and the time and region extraction module to obtain a content portrait of the corresponding content.
[0043] The recommendation model integration module is constructed based on a combination of three models: a collaborative filtering model, a content filtering model, and a deep learning model, and selects corresponding models for processing and analyzing the feature data of the user feature extraction module and the content extraction module to obtain feature data of content that meets the user's interest preferences; in the recommendation model integration module, a collaborative filtering model is adopted based on the user-item rating matrix, and a cosine similarity function is used as a similarity calculation function to measure the similarity between users or items; a content filtering model is adopted based on the feature data obtained in the content feature extraction module, and a support vector machine is used to analyze the user's preference for the content; a deep learning model is adopted based on a convolutional neural network and an LSTM network to construct a deep neural network architecture, and automatically learn the complex feature representations of user feature data and content feature data; and in the model fusion process, the recommendation model integration module adopts an ensemble learning algorithm to perform weighted fusion on the prediction results of multiple models, and the ensemble learning algorithm is stacked generalization;
[0044] The data storage module includes a working data storage module and a knowledge base. The working data storage module is used to store data obtained from each module, and the knowledge base is used to store the vectorized converged media content.
[0045] The regional differentiated recommendation module determines the region to which the user belongs based on the user's latitude and longitude coordinates (x, y), selects a regional feature vector for the corresponding region, and obtains feature data of the recommended content based on the feature data of the content that meets the user's interest preferences obtained by the recommendation model integration module. Then, based on the feature data of the recommended content, the corresponding content is retrieved from the knowledge base for push. The regional feature vector is obtained by vectorizing the data of the geographical location, cultural background, and language habits of the corresponding region.
[0046] In the regional differentiated recommendation module, the region-differentiated recommendation algorithm can be used to perform personalized content recommendation. Combined with the regional division function of the geographic information system (GIS), the cultural, economic, social and other factors of different regions are taken into consideration to construct a region feature vector. At the same time, the spatio-temporal analysis model is used to dynamically match the user's geographic location information and the regional characteristics of the content. According to regional characteristics and user needs, the recommendation results are adjusted and optimized in real time. By introducing the region preference weight factor, the interest preferences of users in different regions are differentiated to ensure that the recommended content not only meets the user's personalized needs but also has distinct regional characteristics, thereby realizing regional differentiated content recommendation services.
[0047] Example 2: A regional differentiated recommendation system for converged media content, comprising a central processing unit, a data collection module, a data vectorization module, a user feature extraction module, a content feature extraction module, a recommendation model integration module, a regional differentiated recommendation module, and a data storage module;
[0048] The central processing unit is used to coordinate the work of various modules;
[0049] The data collection module is used to collect user behavior data;
[0050] The data vectorization module is used to convert data into vector form;
[0051] The user feature extraction module includes a geographic location extraction module and an interest preference module. The geographic location extraction module uses GIS technology to obtain the user's latitude and longitude coordinates (x, y) and stores them in the data storage module. The interest preference module extracts the user's interest preferences and behavior pattern feature data based on the user's behavior data and stores them in the data storage module after processing by the data vectorization module. The user's behavior data includes the user's clicks, browsing, purchases, searches, user's historical browsing records, user's interactive behaviors, and user's basic information. In the interest preference module, it is assumed that the user U's interest preference for category C is P u，c , the calculation formula is as follows: u，c = the number of user U's actions related to category C / the total number of user U's actions;
[0052] The content feature extraction module includes a semantic feature extraction module, an image content feature extraction module, an audio feature extraction module, a time and region extraction module, and a content portrait construction module. The semantic feature extraction module uses a semantic analysis algorithm to process the text content to extract semantic feature data, and stores the extracted semantic feature data in the data storage module after processing by the data vectorization module. The semantic feature data includes semantic feature data of keywords, themes, and emotional tendencies. The semantic feature extraction module uses the following formula:
[0053] Formula I: where n i,j is the number of times word i appears in document j, ∑ k n i,j is the total number of words in document j, TF i,j is the word frequency;
[0054] Formula II: Where N is the total number of documents, n i is the number of documents containing word i, IDF i is the inverse document frequency;
[0055] Formula III (Text Content Feature Extraction): TF-IDF i,j =TF i,j ×IDF i ;
[0056] The image content feature extraction module uses the image recognition model in the field of computer vision to extract features from the picture content, identifies the feature data of objects, scenes, and colors in the image, and stores them in the data storage module after processing by the data vectorization module. The audio feature extraction module uses the audio feature extraction algorithm in the audio processing technology to identify the feature data of the tone, rhythm, and emotion of the audio content, and stores them in the data storage module after processing by the data vectorization module. The time and region extraction module extracts the feature data of the time information and region information of the content release, and stores them in the data storage module after processing by the data vectorization module. The content portrait construction module collects the feature data obtained in the semantic feature extraction module, the image content feature extraction module, the audio feature extraction module, and the time and region extraction module to obtain a content portrait of the corresponding content.
[0057] The recommendation model integration module is constructed based on a combination of three models: a collaborative filtering model, a content filtering model, and a deep learning model. The corresponding model is selected from the feature data of the user feature extraction module and the content extraction module to process and analyze the feature data of the content that meets the user's interest preferences. In the recommendation model integration module, the collaborative filtering model is used based on the user-item rating matrix, and the cosine similarity function is used as the similarity calculation function to measure the similarity between users or items. The content filtering model is used based on the feature data obtained in the content feature extraction module and the logistic regression function is used to analyze the user's preference for the content. The deep learning model is used to construct a deep neural network architecture based on a convolutional neural network and a recurrent neural network to automatically learn the complex feature representation of the user feature data and the content feature data. In addition, during the model fusion process, the recommendation model integration module uses an ensemble learning algorithm to perform weighted fusion on the prediction results of multiple models, and the ensemble learning algorithm is a bagging method.
[0058] The data storage module includes a working data storage module and a knowledge base. The working data storage module is used to store data obtained from each module, and the knowledge base is used to store the vectorized converged media content.
[0059] The regional differentiation recommendation module determines the region to which the user belongs based on the user's latitude and longitude coordinates (x, y), selects the regional feature vector of the corresponding regional region, and obtains the feature data of the recommended content based on the feature data of the content that meets the user's interest preferences obtained by the recommendation model integration module, and then retrieves the corresponding content from the knowledge base based on the feature data of the recommended content for push; the regional feature vector is obtained by vectorizing the data of the geographical location, cultural background, and language habits in the corresponding region.
[0060] Example 3: Figure 1 As shown, the method for using the regional differentiated recommendation system for converged media content described in Example 1 includes the following steps:
[0061] S1. Collecting the converged media content data with authorization and performing deduplication, denoising, formatting, filling in missing data, and standardization or normalization to form a knowledge base, and processing the converged media content through the content feature extraction module to obtain feature data, and vectorizing the feature data and storing it in the knowledge base accordingly;
[0062] S2. With the user's authorization, collect the user's behavior data through a logging system, user tracking technology, and online questionnaires, and process them through the user feature extraction module to obtain the user's latitude and longitude coordinates (x, y) and feature data of the user's interests, preferences, and behavior patterns;
[0063] S3, processing and analyzing the characteristic data of the user's interest preferences and behavior patterns obtained in step S2 by the recommendation model integration module to obtain characteristic data of content that meets the user's interest preferences;
[0064] S4. Determine the region to which the user belongs based on the user's latitude and longitude coordinates (x, y) through the regional differentiation recommendation module, select the regional feature vector of the corresponding regional region, and obtain the feature data of the recommended content based on the feature data of the content that meets the user's interest preferences obtained by the recommendation model integration module. Then, according to the feature data of the recommended content, retrieve the corresponding content from the knowledge base for push, that is, display the retrieved integrated media content to the user in the form of a recommendation bar, pop-up window, or personalized homepage, and provide a user feedback mechanism to collect user satisfaction with the recommendation results and improvement suggestions.
[0065] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.
Claims
1. A regional differentiated recommendation system for converged media content, characterized by: The system includes a central processing unit, a data collection module, a data vectorization module, a user feature extraction module, a content feature extraction module, a recommendation model integration module, a regional differentiation recommendation module, and a data storage module; The central processing unit is used to coordinate the work of various modules; The data collection module is used to collect user behavior data; The data vectorization module is used to convert data into vector form; The user feature extraction module includes a geographic location extraction module and an interest preference module. The geographic location extraction module uses GIS technology to obtain the user's latitude and longitude coordinates (x, y) and stores them in the data storage module. The interest preference module extracts the user's interest preferences and behavior pattern feature data based on the user's behavior data and stores the feature data in the data storage module after processing by the data vectorization module. The content feature extraction module includes a semantic feature extraction module, an image content feature extraction module, an audio feature extraction module, a time and region extraction module, and a content portrait construction module. The semantic feature extraction module uses a semantic analysis algorithm to process the text content to extract semantic feature data and stores it in the data storage module after processing by the data vectorization module. The image content feature extraction module uses an image recognition model in the field of computer vision to extract features of the picture content, identify the feature data of objects, scenes, and colors in the image, and stores it in the data storage module after processing by the data vectorization module. The audio feature extraction module uses an audio feature extraction algorithm in audio processing technology to identify the feature data of the tone, rhythm, and emotion of the audio content and stores it in the data storage module after processing by the data vectorization module. The time and region extraction module extracts the feature data of the time information and region information of the content release and stores it in the data storage module after processing by the data vectorization module. The content portrait construction module collects the feature data obtained in the semantic feature extraction module, image content feature extraction module, audio feature extraction module, and time and region extraction module to obtain a content portrait of the corresponding content. The recommendation model integration module is constructed based on a combination of one or more models selected from the collaborative filtering model, the content filtering model, and the deep learning model, and selects corresponding models to process and analyze the feature data of the user feature extraction module and the content extraction module to obtain feature data of content that meets the user's interest preferences; The data storage module includes a working data storage module and a knowledge base. The working data storage module is used to store data obtained from each module, and the knowledge base is used to store the vectorized converged media content. The regional differentiated recommendation module determines the region to which the user belongs based on the user's latitude and longitude coordinates (x, y), selects the regional feature vector of the corresponding region, and obtains the feature data of the recommended content based on the feature data of the content that meets the user's interest preferences obtained by the recommendation model integration module, and then retrieves the corresponding content from the knowledge base based on the feature data of the recommended content for push.
2. The regional differentiated recommendation system for converged media content according to claim 1 is characterized by: The user's behavior data includes the user's clicks, browsing, purchases, searches, historical browsing records, user interaction behaviors, and basic information of the user.
3. The regional differentiated recommendation system for converged media content according to claim 1 is characterized by: In the interest preference module, it is assumed that user U’s interest preference for category C is P u，c , the calculation formula is as follows: u，c =Number of user U's behaviors related to category C / total number of user U's behaviors.
4. The regional differentiated recommendation system for converged media content according to claim 1 is characterized by: The semantic feature data includes semantic feature data of keywords, topics and emotional tendencies; the semantic feature extraction module adopts the following formula: Formula I: where n i,j is the number of times word i appears in document j, ∑ k n i,j is the total number of words in document j, TF i,j is the word frequency; Formula II: Where N is the total number of documents, n i is the number of documents containing word i, IDF i is the inverse document frequency; Formula III: TF-IDF i,j =TF i,j ×IDF i .
5. The regional differentiated recommendation system for converged media content according to claim 1 is characterized by: In the recommendation model integration module, a collaborative filtering model is adopted based on the user-item rating matrix, and a similarity calculation function is used to measure the similarity between users or items. A content filtering model is adopted based on the feature data obtained in the content feature extraction module, and a classification algorithm or regression function is used to analyze the user's preference for the content. A deep learning model such as a convolutional neural network, a recurrent neural network and its variants is adopted to construct a deep neural network architecture to automatically learn the complex feature representations of user feature data and content feature data; and in the model fusion process, the recommendation model integration module adopts an integrated learning algorithm to perform weighted fusion on the prediction results of multiple models, and the integrated learning algorithm includes stacking generalization, bagging method and boosting method.
6. The regional differentiated recommendation system for converged media content according to claim 1 is characterized by: The regional feature vector is obtained by vectorizing the data of the geographical location, cultural background, and language habits in the corresponding region.
7. The method for using the regional differentiated recommendation system for converged media content according to any one of claims 1 to 6, characterized in that: The following steps are involved: S1. Collecting converged media content data, removing duplication, removing noise, unifying the format, filling in missing data, and standardizing or normalizing the data to form a knowledge base, and processing the converged media content through the content feature extraction module to obtain feature data, and vectorizing the feature data and storing it in the knowledge base accordingly; S2. With the user's authorization, collect the user's behavior data through a logging system, user tracking technology, and online questionnaires, and process them through the user feature extraction module to obtain the user's latitude and longitude coordinates (x, y) and feature data of the user's interests, preferences, and behavior patterns; S3, processing and analyzing the characteristic data of the user's interest preferences and behavior patterns obtained in step S2 by the recommendation model integration module to obtain characteristic data of content that meets the user's interest preferences; S4. Determine the region to which the user belongs based on the user's latitude and longitude coordinates (x, y) through the regional differentiation recommendation module, select the regional feature vector of the corresponding regional area, and obtain the feature data of the recommended content based on the feature data of the content that meets the user's interest preferences obtained by the recommendation model integration module, and then retrieve the corresponding content from the knowledge base based on the feature data of the recommended content for push.
8. The method of use according to claim 7, characterized in that: In step S4, the retrieved integrated media content is displayed to the user in the form of a recommendation column, pop-up window, and personalized homepage, and a user feedback mechanism is provided to collect user satisfaction with the recommendation results and improvement suggestions.