Media service intelligent recommendation method and device, equipment and medium

Abstract

The invention relates to a media service intelligent recommendation method and device, equipment and a medium. According to the method, semantic analysis and emotional tendency analysis are carried out on a news comment text through a pre-established semantic map and an emotional feature library to generate a compliance evaluation result, semantic reconstruction and emotional adjustment are carried out on the text based on the evaluation result to generate an optimized text, and multi-terminal adaptive real-time distribution is realized by utilizing an intelligent streaming media architecture. User feedback data is combined to analyze interaction behaviors and preference characteristics, user behaviors are mapped to a virtual scene through a digital twinning technology, experience data are collected, and content recommendation parameters and push logic are dynamically adjusted based on emotion response analysis to form an all-media propagation matrix configuration scheme. And finally, continuously optimizing the propagation strategy through operation data analysis, thereby realizing the technical effects of news content compliance accurate control, multi-terminal propagation efficiency improvement, user demand dynamic perception and propagation strategy adaptive optimization.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence and natural language processing, and in particular relates to a method, device, equipment and medium for intelligent recommendation of media services. Background Art

[0002] In today's information age, media services, as the core hub for information dissemination and social connection, play an irreplaceable and important role in cultural heritage, public opinion guidance, and smart city development. With the rapid development of digital technology, intelligent recommendation and omnimedia communication have become key areas for improving user experience and communication efficiency. However, existing methods often expose significant shortcomings when faced with complex social needs and diverse user scenarios.

[0003] Traditional news comment management relies heavily on simple keyword filtering, which struggles to adapt to the diverse cultural contexts and emotional expressions. This single-minded approach not only easily misjudges legitimate content but also filters out expressions with cultural connotations, limiting dissemination effectiveness. Furthermore, conventional dissemination methods are limited to a one-way and linear model, lacking interactivity and immersion, and failing to meet users' diverse content needs.

[0004] When it comes to personalized recommendations, existing systems lack multi-dimensional data integration and dynamic behavior perception, making it difficult to capture evolving user interests and resulting in inaccurate and relatable recommendations. Furthermore, fragmented technical architectures hinder cross-platform services and full-scenario adaptation. Existing solutions often struggle to meet the demands of multi-terminal deployment and efficient interaction within a self-controlled environment.

[0005] Therefore, how to ensure content compliance and cultural adaptation while improving recommendation accuracy through multi-dimensional data integration and building a technical architecture that supports full-scene adaptation and efficient interaction has become a key issue in promoting the intelligence of media services. Summary of the Invention

[0006] Based on this, it is necessary to provide a method, device, equipment and medium for intelligent recommendation of media services to address the above technical issues.

[0007] In a first aspect, the present application provides a method for intelligently recommending media services, comprising:

[0008] S1. Perform semantic parsing and sentiment analysis on the input news commentary text using a pre-established semantic graph and sentiment feature library to generate preliminary compliance assessment results.

[0009] S2. Optimize the news commentary text based on the preliminary compliance assessment results to generate optimized text that meets compliance standards;

[0010] S3, embed optimized text into the dissemination process through intelligent streaming media architecture and perform multi-terminal linkage distribution operations;

[0011] S4. Obtain user feedback data based on the results of the multi-terminal linkage distribution operation;

[0012] S5. Analyze user interaction behavior based on user feedback data to generate user acceptance judgment results;

[0013] S6. Use digital twin technology to map user interaction behaviors to virtual scenes and generate user experience data;

[0014] S7. Analyze user emotional responses based on user experience data to obtain analysis results; adjust content recommendation parameters and optimize push logic based on the analysis results to generate a full-media communication matrix configuration plan;

[0015] S8. Analyze the operating data of the full-media communication matrix configuration plan based on the preset performance indicators to determine the communication effectiveness and user engagement; optimize the semantic map and emotional feature library based on the communication effectiveness and user engagement.

[0016] In a second aspect, the present application further provides a media service intelligent recommendation device, comprising:

[0017] The semantic parsing and sentiment analysis module is used to perform semantic parsing and sentiment analysis on the input news commentary text using a pre-established semantic graph and sentiment feature library, generating preliminary compliance assessment results.

[0018] A text optimization processing module is used to optimize news commentary text based on preliminary compliance assessment results to generate optimized text that meets compliance standards;

[0019] The text embedding and distribution module is used to embed optimized text into the dissemination process through the intelligent streaming media architecture and perform multi-terminal linkage distribution operations;

[0020] User feedback data acquisition module, used to obtain user feedback data based on the results of multi-terminal linkage distribution operations;

[0021] User interaction behavior analysis module, used to analyze user interaction behavior based on user feedback data and generate user acceptance judgment results;

[0022] The virtual scene mapping module is used to map user interaction behaviors to virtual scenes through digital twin technology to generate user experience data;

[0023] The emotional response analysis module is used to analyze user emotional responses based on user experience data and obtain analysis results;

[0024] The recommendation optimization module is used to adjust content recommendation parameters and optimize push logic based on the analysis results, and generate a full-media communication matrix configuration plan;

[0025] The communication effectiveness analysis module is used to analyze the operating data of the full-media communication matrix configuration plan based on preset effectiveness indicators to determine the communication effectiveness and user participation;

[0026] The database optimization module is used to optimize the semantic graph and sentiment feature library based on communication effectiveness and user engagement.

[0027] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements a media service intelligent recommendation method as in the first aspect.

[0028] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a media service intelligent recommendation method as in the first aspect.

[0029] The above-mentioned media service intelligent recommendation method, device, equipment and medium perform semantic analysis and sentiment tendency analysis on news comment text through a pre-established semantic map and sentiment feature library to generate compliance assessment results, perform semantic reconstruction and sentiment adjustment on the text based on the assessment results to generate optimized text, use intelligent streaming media architecture to achieve real-time distribution of multi-terminal adaptation, combine user feedback data to analyze interactive behavior and preference characteristics, map user behavior to virtual scenes and collect experience data through digital twin technology, dynamically adjust content recommendation parameters and push logic based on sentiment response analysis to form an all-media communication matrix configuration plan, and finally continuously optimize the communication strategy through operation data analysis to achieve the technical effects of precise control of news content compliance, improved multi-terminal communication efficiency, dynamic perception of user needs and adaptive optimization of communication strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0031] Figure 1 A flowchart of a media service intelligent recommendation method provided by the present invention;

[0032] Figure 2 A schematic diagram of a process for generating user experience data in an optional embodiment of the present invention;

[0033] Figure 3 This is a structural diagram of a media service intelligent recommendation device provided by the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0035] refer to Figure 1 , which presents a flow chart of a media service intelligent recommendation method provided by this application, the method comprising the following steps:

[0036] S1. Perform semantic parsing and sentiment analysis on the input news comment text through the pre-established semantic graph and sentiment feature library to generate preliminary compliance assessment results.

[0037] Specifically, a value semantic map is first constructed, encompassing multiple value nodes and establishing relationships between them. Simultaneously, a sentiment language feature library is established, collecting a large number of sentiment-laden words and phrases, including positive sentiment words (such as "excellent," "outstanding," and "wonderful"), negative sentiment words (such as "bad," "poor," and "angry"), and neutral sentiment words, annotated with corresponding sentiment values ​​and intensity levels. Upon inputting news commentary text, the natural language processing module of the multimodal large model is invoked to perform lexical analysis, syntactic analysis, and semantic role labeling on the text, extracting key semantic units. These extracted semantic units are then matched against the value semantic map, calculating the semantic similarity between the text and each core value node to determine the value orientation conveyed by the text. Simultaneously, a sentiment analysis algorithm, combined with the sentiment feature library, is used to determine the sentiment orientation of the words in the text. The distribution and weighting of positive, negative, and neutral sentiment words are statistically analyzed, and the overall sentiment orientation value and intensity level are calculated, generating preliminary compliance assessment results, which may include indicators such as value conformance and sentiment orientation compliance.

[0038] S2. Optimize the news commentary text based on the preliminary compliance assessment results to generate optimized text that meets compliance standards.

[0039] Specifically, when the preliminary evaluation results show that the text has a value deviation or an emotional tendency that does not comply with regulations, the text optimization mechanism is triggered. On the one hand, for value deviations, the value concepts that should be emphasized in the text can be determined based on the value nodes and their associations in the semantic map, and supplementary sentences or correction suggestions can be generated through techniques such as synonym expansion and semantic association. On the other hand, for emotional tendencies that do not comply with regulations, if the emotions are too negative or too intense, the emotional vocabulary in the text can be replaced or modified through the emotional adjustment algorithm, such as replacing "anger" with "dissatisfaction" and replacing "very bad" with "needs to be improved". At the same time, the sentence structure can be adjusted to make its expression more peaceful and objective. The optimized text will be re-analyzed for semantics and emotional tendency until it meets the compliance standards, that is, the value compliance reaches the preset threshold and the emotional tendency is within a reasonable range.

[0040] S3: Embed optimized text into the dissemination process through intelligent streaming media architecture and perform multi-terminal linkage distribution operations.

[0041] Specifically, the intelligent streaming media architecture is designed based on a microservices architecture and includes multiple sub-modules such as content acquisition services, content processing services, and content distribution services. The sub-modules interact with each other through lightweight communication protocols to achieve a highly cohesive and low-coupling system architecture. The optimized text content is first encapsulated into streaming media data packets that conform to a specific format, containing text content, metadata (such as release time, author information, content theme, etc.), and associated multimedia resource links (such as pictures, videos, etc.). Then, according to the preset dissemination strategy, the streaming media data packets are sent to different distribution service nodes, which correspond to multiple carriers such as mobile applications, web platforms, smart TVs, and social media platforms. During the distribution process, the content delivery network (CDN) technology is used, combined with real-time network traffic monitoring and node load balancing algorithms, to ensure that the optimized text can be transmitted quickly and stably to the user devices at each end, realizing multi-terminal linkage distribution, that is, displaying relevant content simultaneously on different terminals. At the same time, the text content is adaptively adjusted according to the user characteristics and display characteristics of each terminal, such as optimizing the layout on the mobile terminal to adapt to small-screen reading, and displaying large subtitles and high-definition images on the smart TV terminal.

[0042] S4. Based on the results of the multi-terminal linkage distribution operation, obtain user feedback data.

[0043] Specifically, a user behavior monitoring module is integrated into the client application of each end carrier. This module uses event monitoring technology to capture the user's interactive behavior with news comment content in real time, including click operations, reading time, comment behavior (such as posting comments, liking, sharing, etc.), scrolling behavior (such as the frequency and depth of sliding up and down to view content), and collection behavior of specific content. These user behavior data will be timestamped and device identified, and sent to the user feedback data warehouse of the back-end server through a secure data transmission channel for storage. At the same time, in order to obtain more comprehensive user feedback, a user questionnaire entrance can be set up on the client to regularly collect users' subjective evaluation data on content quality, relevance, attractiveness, etc. In addition, the open API of the social media platform can be used to obtain users' social behavior data such as forwarding, discussion, and liking of related content on social media, and integrate it into the user feedback data system to understand users' acceptance and feedback on optimized text from multiple dimensions.

[0044] S5. Analyze user interaction behavior based on user feedback data to generate user acceptance judgment results.

[0045] Specifically, data mining and machine learning algorithms can be used to conduct in-depth analysis of user feedback data. First, association rule mining algorithms are used to analyze the relationship between user reading time and commenting behavior. For example, users who read longer are more likely to leave comments, thus concluding that reading depth and user engagement are correlated. Second, cluster analysis algorithms are used to categorize user groups based on their behavioral characteristics, such as "deep readers and active commenters," "occasional browsers," and "sharers." Based on the behavioral patterns of different user categories, corresponding user acceptance evaluation indicators are developed. For users who read deeply and actively comment, the focus is on indicators such as comment quality and number of likes; for occasional browsers, the focus is on the distribution of reading time and browsing frequency; and for sharers, the focus is on analyzing their sharing platform preferences and the impact of their sharing. By calculating weighted scores for each indicator and combining classification algorithms (such as decision trees and support vector machines), user behavior categories are determined. Ultimately, user acceptance results are generated, which can include quantitative indicators such as user satisfaction, interest, and engagement with the content, as well as descriptions of the distribution and behavioral characteristics of different user groups, providing a basis for subsequent adjustments to content recommendation strategies.

[0046] S6. Use digital twin technology to map user interaction behaviors to virtual scenes and generate user experience data.

[0047] Specifically, 3D modeling software can be combined with Geographic Information System (GIS) data to construct a virtual urban space that is highly consistent with the real scene, including elements such as buildings, roads, and public facilities, and each element is given a unique identifier and attribute information. At the same time, a multimodal interaction framework is developed to integrate hardware facilities such as motion capture equipment, gesture recognition sensors, and voice interaction modules, as well as corresponding software algorithms, to enable users to interact naturally in virtual scenes. The user interaction behavior data is mapped with the elements in the digital twin scene. For example, when a user likes a news comment about construction progress on a news client, a virtual "like" gesture is generated in the digital twin scene and appears near the building or project area associated with the news content; the user's sharing behavior is mapped to the transmission of a virtual information package containing news content to other virtual users (representing their social friends) in the virtual scene. Through a real-time rendering engine, these user interactions are dynamically presented in the digital twin scene. Simultaneously, information such as user movement trajectory, dwell time, and interaction frequency within the virtual scene is recorded to generate user experience data. This data can include indicators of user immersion in the virtual space (such as visual focus distribution and the smoothness of virtual movements), interaction efficiency (such as the time required to complete specific interactive tasks), and attention paid to different elements in the virtual scene. This step can fully reflect the user's real experience during media content consumption, providing intuitive and vivid data support for optimizing media services.

[0048] S7. Analyze user emotional responses based on user experience data to obtain analysis results; adjust content recommendation parameters and optimize push logic based on the analysis results to generate an all-media communication matrix configuration plan.

[0049] Specifically, an emotional response model is first established. This model can be based on the recurrent neural network (RNN) or long short-term memory (LSTM) architecture used in deep learning to perform serialized analysis of user experience data. Various indicators from the user experience data (such as immersion, interaction efficiency, and attention) are used as model input features. Combined with basic user attributes (such as age, gender, occupation) and historical behavioral data, the model is trained to learn user emotional response patterns. The model predicts users' emotional states in different media content consumption scenarios, including emotion categories such as pleasure, curiosity, boredom, and anxiety, and calculates the corresponding emotion intensity values ​​to obtain user emotional response analysis results. Based on the analysis results, content recommendation parameters are adjusted. For example, if users' emotional states are mostly pleasant and of high intensity when experiencing a certain type of technology news content, the weight of this type of content in the recommendation algorithm is increased; if users express boredom with a certain type of entertainment news content, its recommendation frequency is reduced. At the same time, we optimize push logic, determining the optimal time, frequency, and format for push notifications based on the user's emotional state and behavioral habits. For example, for users experiencing anxiety, we avoid using overly stimulating titles and images when pushing content, and appropriately reduce the frequency of push notifications to avoid burdening the user. Based on the above adjustments and optimizations, we generate a full-media communication matrix configuration plan. This plan covers content delivery strategies, push schedules, and collaborative linkage rules across different media platforms (such as news clients, social media, smart TVs, and outdoor electronic screens). This ensures that media content is pushed to the most interested user groups at the right time and in the right format, maximizing the communication effect.

[0050] S8. Analyze the operating data of the full-media communication matrix configuration plan based on the preset performance indicators to determine the communication effectiveness and user engagement; optimize the semantic map and emotional feature library based on the communication effectiveness and user engagement.

[0051] Specifically, the preset performance indicators may include the scope of content dissemination (such as the number of readings, playbacks, and forwardings), the depth of content dissemination (such as the number of user comments, likes, and shares), user retention rate (such as the number of new user registrations, the return visit rate of old users, etc.), and content conversion rate (such as the purchase behavior and registration for activities generated by users due to content guidance). The data collection system collects relevant operating data after the implementation of the full-media communication matrix configuration plan, such as traffic statistics, user behavior logs, and business conversion records of each platform, and compares and analyzes them with the preset performance indicators. Use statistical analysis methods to calculate the completion of each performance indicator, such as calculating the ratio of actual reading volume to the expected target reading volume to evaluate whether the content dissemination scope has met expectations; analyze the proportion of user comments to the total reading volume to determine whether the content dissemination depth meets the requirements.

[0052] Based on the analysis results of the effectiveness indicators, the communication effectiveness and user engagement are judged. If the communication effectiveness and user engagement do not meet the expected goals, the optimization process of the semantic graph and emotional feature library is initiated. For the semantic graph, the corresponding semantic nodes and associations are supplemented based on the common differences in value expression and semantic understanding deviations in user feedback. For the emotional feature library, based on the unrecognized or misrecognized emotional words that appear in the user's emotional response analysis results, the emotional feature library is updated in a timely manner, new emotional words and their emotional tendency values ​​and intensity levels are added, or the annotation information of existing emotional words is corrected. This allows the semantic graph and emotional feature library to better adapt to the user's actual language expression habits and emotional expression methods, and realize a closed-loop management mechanism for continuous optimization and improvement.

[0053] The above-mentioned intelligent recommendation method for media services performs semantic analysis and sentiment tendency analysis on news commentary texts through a pre-established semantic map and sentiment feature library to generate compliance assessment results, performs semantic reconstruction and sentiment adjustment on the text based on the assessment results to generate optimized text, utilizes an intelligent streaming media architecture to achieve real-time distribution of multi-terminal adaptation, combines user feedback data to analyze interactive behaviors and preference characteristics, maps user behaviors to virtual scenes and collects experience data through digital twin technology, dynamically adjusts content recommendation parameters and push logic based on sentiment response analysis to form an all-media communication matrix configuration plan, and finally continuously optimizes the communication strategy through operational data analysis to achieve the technical effects of precise control of news content compliance, improved multi-terminal communication efficiency, dynamic perception of user needs, and adaptive optimization of communication strategies.

[0054] In an optional embodiment, S1 includes the following steps:

[0055] S11. Extract core semantic features from the semantic graph, perform sentiment analysis on the news comment text based on the sentiment feature library, and obtain the sentiment tendency.

[0056] Specifically, graph neural network algorithms can be used to mine node importance and relevance, extracting key node feature vectors that represent the text's themes and values. Word embedding technology and sentiment analysis algorithms can be used to map text into a sentiment vector space, calculating similarity with words in a sentiment feature library to derive the text's sentiment.

[0057] S12. Based on the core semantic features and emotional tendencies, the news commentary text is subjected to dual analysis of semantic content and emotional expression to determine the degree of semantic deviation and emotional tendency value.

[0058] Specifically, semantic content analysis compares the text's semantics with core semantic features to determine the degree of deviation. Sentiment analysis quantifies the strength and direction of sentiment. The degree of semantic deviation is determined by calculating the semantic similarity between the text and a standard semantic library; the lower the similarity, the greater the deviation. Sentimental tendency is calculated using a sentiment analysis algorithm. The model identifies sentiment vocabulary in the text and combines it with annotation information to evaluate the sentiment tendency.

[0059] S13. If the degree of semantic deviation or the sentiment tendency value exceeds the preset threshold, the news comment text is judged to be non-compliant with the compliance standards.

[0060] Specifically, the preset threshold is the standard for determining whether a text is compliant. The threshold setting can be based on historical data and expert experience. When determining the compliance of a text, the calculated degree of semantic deviation and sentiment value are compared with it. If the degree of semantic deviation exceeds the threshold, it indicates that there is a significant difference between the semantics of the text and the standard; if the sentiment value exceeds the threshold range, it indicates that the sentiment of the text is too extreme and does not comply with the regulations. In this case, the text is marked as non-compliant, triggering the subsequent optimization process.

[0061] S14. Comprehensively evaluate the degree of semantic deviation and sentiment tendency value through a multimodal large language model to generate preliminary compliance assessment results.

[0062] Specifically, the multimodal large language model integrates multimodal data such as text, images, and audio to comprehensively understand text semantics and sentiment. Based on a deep learning architecture and trained on large amounts of text and annotated data, the model can accurately identify text features and patterns. During comprehensive evaluation, the model takes the degree of semantic deviation and sentiment as input. Through calculation and reasoning, it generates an assessment report that includes a text compliance score, a detailed description of semantic deviation, and a sentiment analysis.

[0063] In an optional embodiment, S2 includes the following steps:

[0064] S21. When the preliminary compliance assessment results show that there is a semantic deviation or the emotional expression does not meet the preset threshold, the news comment text is semantically reconstructed.

[0065] Specifically, the core of semantic reconstruction lies in optimizing text using the structured knowledge of semantic graphs. Through techniques such as synonym replacement, semantic association expansion, and semantic relationship reorganization, semantic reconstruction suggestions are generated to create semantic expressions that better align with the target values.

[0066] S22. Adjust the emotional expression of the news commentary text to generate an adjusted text that meets preset emotional characteristics.

[0067] Specifically, the core of emotion adjustment is to optimize the emotional expression in the text through the emotion adjustment algorithm. If the text emotion is too intense or tends to be negative, it can be adjusted through technical means such as word replacement, emotion intensity adjustment, and emotion tendency guidance. For example, if there is anger or dissatisfaction in the text, replace the overly intense words (such as "anger" with "dissatisfaction"), and adjust the sentence structure and tone to make the text expression more peaceful and objective. It is also possible to generate an emotion expression template that meets the requirements based on the preset emotion features in the emotion feature library for users to refer to or directly apply, to ensure that the emotional expression of the text meets the preset standards.

[0068] S23. Conduct a compliance review on the adjusted text after semantic reconstruction and sentiment adjustment to obtain the review result.

[0069] Specifically, the review process uses a multimodal large language model, combined with a semantic map and sentiment feature library, to conduct a comprehensive semantic and sentiment evaluation of the text. The multimodal large language model can not only understand the semantic content of the text, but also combine multimodal data such as images and audio to make a comprehensive judgment on the emotional expression and semantic background of the text. The text is evaluated in multiple dimensions through methods such as semantic similarity calculation, sentiment tendency analysis, and semantic deviation detection. Semantic similarity calculation is used to measure the degree of match between the text and the value semantic map; sentiment tendency analysis is used to determine whether the emotional expression of the text conforms to the preset sentiment feature library; semantic deviation detection is used to identify whether the text contains semantic content that contradicts mainstream values. Finally, the evaluation results of each dimension are combined to generate a detailed review report, including the text's semantic deviation level, sentiment tendency value, and overall compliance score, to ensure that the text meets the specified standards in terms of both semantics and sentiment.

[0070] S24. If the review result does not meet the standard, the semantic reconstruction and sentiment adjustment operations are repeated until the review result meets the standard and an optimized text is obtained.

[0071] Specifically, an iterative optimization mechanism is used to ensure that the text gradually approaches compliance requirements. During each iteration, the text modification history and evaluation results are recorded, and the effectiveness of the optimization path is analyzed through machine learning algorithms. If the text still fails to meet the compliance standards after multiple iterations, the expert review mechanism can be triggered, and manual intervention can be performed to evaluate the text and provide modification suggestions. At the same time, the features of the text that failed are fed back to the semantic graph and sentiment feature library to optimize the content of the knowledge base and the parameters of the evaluation model, thereby improving the overall performance and accuracy of the system. The optimized text not only meets the prescribed standards in terms of semantics and sentiment, but also maintains the main points and emotional expression characteristics of the original comment text, ensuring the authenticity and coherence of the content. The final optimized text will be stored in the content library for subsequent dissemination and recommendation.

[0072] refer to Figure 2 In an optional embodiment, S6 includes the following steps:

[0073] S61. Based on the interactive behaviors in user feedback data, digital twin technology is used to build a user behavior model.

[0074] Specifically, user feedback data covers various interactive behaviors of users on multiple terminals, such as clicks, reading, comments, and sharing. Through data acquisition, these behavioral data are collected in real time and pre-processed, including data cleaning, denoising, format unification and other operations. Using digital twin technology, user behavior is first digitally modeled. When constructing a user behavior model, the user's basic attributes (such as age, gender, occupation, etc.) and historical behavior records are combined, and machine learning algorithms (such as hidden Markov models, deep neural networks, etc.) are used to analyze and learn user behavior patterns to generate a model that includes user behavior feature vectors and behavior probability distributions. This model dynamically reflects the user's behavioral habits and preferences in the process of media content consumption.

[0075] S62. Map the user behavior model to the virtual scene and simulate the immersive news narrative space through the real-time rendering engine.

[0076] Specifically, based on the constructed digital twin scene, the behavioral feature vectors in the user behavior model are associated and mapped with elements in the virtual scene. For example, a user's browsing behavior is mapped to the user's movement trajectory in the virtual scene, and comment behavior is mapped to the user's voice or text bubbles in the virtual scene. The real-time rendering engine combines the physics engine and the multimodal interaction framework to render and simulate the performance of user behavior in the virtual scene in real time. The physics engine ensures that the movement and interaction of objects in the virtual scene conform to the laws of real physics, such as the feedback force and motion effects generated when a user clicks on a virtual object. The multimodal interaction framework supports multiple user interaction methods, such as gesture recognition, voice control, and eye contact, allowing users to operate and experience the virtual scene naturally. Through the above technical means, a highly immersive virtual news narrative space that is closely related to user behavior is constructed.

[0077] S63. In the immersive news narrative space, obtain the user's operation data and reaction data in the virtual and real environment.

[0078] Specifically, as users operate in a virtual scene, integrated motion capture devices, sensors, and monitoring modules capture detailed user data in real time, including gestures, movement trajectories, and interaction frequency. Simultaneously, biometric monitoring devices (such as eye trackers and heart rate sensors) and sentiment analysis algorithms within the virtual scene capture user responses within the virtual environment, such as visual focus distribution, emotional fluctuations, and dwell time. This data is then sent in real time to a backend server via a data transmission protocol for storage and processing.

[0079] S64. Generate user experience data based on the operation data and response data.

[0080] Specifically, operational data analysis focuses on the efficiency and fluency of user interactions within virtual scenes. This can be assessed through metrics like the time required to complete specific tasks and the degree of simplification of operational steps. Reaction data analysis focuses on users' emotional experience and sense of immersion. Emotional recognition algorithms can be used to categorize and quantify user emotional fluctuations, while metrics like visual focus distribution and dwell time are combined to assess users' attention and interest in different virtual elements. The resulting user experience data includes quantitative indicators across multiple dimensions, including overall user satisfaction within the virtual scene, immersion ratings, and the types and intensity of emotional experiences. This provides a concrete and intuitive basis for optimizing media communication strategies and service quality.

[0081] In an optional embodiment, the step of “adjusting content recommendation parameters and optimizing push logic based on the analysis results to generate an all-media communication matrix configuration plan” includes the following steps:

[0082] S71. Calculate the adjustment amount of the content recommendation parameter based on the emotional response characteristics in the analysis result to obtain the adjusted content recommendation parameter.

[0083] Specifically, emotional response features are key information extracted from user experience data, encompassing the emotional state and intensity of users in different media content consumption scenarios. First, the emotional response features are quantified and converted into computable emotional feature vectors. Then, a pre-defined emotion-to-recommendation parameter mapping model, trained based on historical data and machine learning algorithms, is used to reflect the complex relationship between emotional features and recommendation parameters. Through model calculations, adjustments to content recommendation parameters are derived, including adjustments to recommendation weight, frequency, and order. For example, if a user's emotional response features indicate high pleasure and interest in a particular type of content, the recommendation weight and frequency for that content are increased accordingly. Conversely, if the emotional response features indicate boredom or aversion, the recommendation parameters are decreased. The original recommendation parameters can be combined with the adjustments through matrix operations or optimization algorithms to generate adjusted content recommendation parameters.

[0084] S72. Based on the preset push logic optimization rules, the timeliness weight and topic relevance coefficient in the push logic are dynamically updated to obtain the updated push logic.

[0085] Specifically, push logic optimization rules can be formulated based on the timeliness and topic relevance of content dissemination, aiming to ensure that users receive highly relevant content at the appropriate time. Content timeliness metrics can be monitored in real time, such as the time of news events, content update frequency, and user preferences for time-sensitive content. Based on these metrics, the timeliness weight can be dynamically adjusted using a preset timeliness decay function or reinforcement rules. For example, for breaking news events, the timeliness weight is initially high, but over time, the weight gradually decreases according to a preset decay function. To update the topic relevance coefficient, by analyzing the user's historical behavior data and current interests, combined with a semantic analysis algorithm, the semantic similarity and correlation strength between the content and the user's topics of interest are calculated. The topic relevance coefficient is updated based on a preset relevance threshold and adjustment step size. For example, if a user has recently frequently followed a topic and new content has a strong correlation with that topic, the topic relevance coefficient can be increased to increase the likelihood of relevant content being pushed.

[0086] S73: Combine and match the adjusted content recommendation parameters with the updated push logic to generate a set of candidate configuration solutions.

[0087] Specifically, by constructing a multidimensional parameter space, content recommendation parameters (such as recommendation weight, frequency, and order) and push logic parameters (such as timeliness weight and topic relevance coefficient) are combined and arranged to generate a large number of candidate configurations. To improve efficiency, heuristic search algorithms or genetic algorithms can be used to search for potential high-quality configurations in the parameter space. Each candidate configuration contains a specific set of content recommendation parameters and push logic parameters, which can define a unique omnimedia communication strategy. These candidate solutions are initially screened to eliminate solutions that clearly do not meet basic communication requirements or have low performance, and retain a certain number of candidate solutions for the next evaluation step. During the screening process, simple rule filtering or quick evaluation based on predefined performance indicators can be used, such as checking the diversity of recommended content and the rationality of push timing.

[0088] S74. Predict the communication effects of the candidate configuration scheme set through Monte Carlo simulation, and select the configuration scheme with the best predicted value as the full-media communication matrix configuration scheme.

[0089] Specifically, Monte Carlo simulation is a numerical calculation method based on random sampling and probabilistic statistics that effectively handles uncertainty and complexity in communication effect prediction. A communication effect prediction model is established for each candidate configuration, comprehensively considering multiple variables such as user behavior characteristics, content dissemination characteristics, and network environment factors. Through Monte Carlo simulation, each candidate is randomly sampled and simulated multiple times, generating a large number of possible dissemination scenarios and result data. The simulation results are statistically analyzed to calculate dissemination effect metrics for each candidate, such as content reach, user engagement, and conversion rate. Based on pre-defined optimization objectives (such as maximizing reach and user engagement), the candidate options are ranked and screened using multi-objective optimization algorithms or decision tree methods. Ultimately, the configuration with the best predicted dissemination effect is selected as the all-media dissemination matrix configuration. This configuration will be applied in actual media dissemination to maximize content dissemination effects and increase user engagement.

[0090] The above-mentioned intelligent recommendation method for media services generates compliance assessment results by constructing a semantic graph and an emotional feature library to perform multimodal semantic analysis and emotional tendency analysis on news commentary texts. Based on the assessment results, the multimodal model is used to implement semantic reconstruction and emotional feature adjustment of the text to generate optimized content. The intelligent streaming media architecture is used to realize multi-terminal adaptive real-time distribution and communication effect monitoring. The interactive behavior is analyzed and preference portraits are constructed based on user feedback data. Digital twin technology is used to map user behavior to virtual scenes to form virtual-real experience data. The emotional response characteristics are extracted through the multimodal interaction framework and the recommendation parameters and push logic are dynamically adjusted. Finally, based on Monte Carlo simulation, an all-media communication matrix configuration plan is generated and the semantic matching rules are continuously optimized to realize intelligent compliance review of the entire process of news content production and dissemination, precise distribution across platforms, dynamic perception of user needs, and adaptive iterative optimization of communication strategies, effectively improving content dissemination efficiency and user participation.

[0091] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0092] Based on the same inventive concept, embodiments of the present application also provide an apparatus for implementing the aforementioned method for intelligent media service recommendation. The solution provided by this apparatus is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the apparatus for intelligent media service recommendation provided below can be found in the aforementioned limitations of the method for intelligent media service recommendation, and will not be further elaborated here.

[0093] In an exemplary embodiment, Figure 3 As shown, a media service intelligent recommendation device 300 is provided, comprising:

[0094] The semantic parsing and sentiment analysis module 301 is used to perform semantic parsing and sentiment tendency analysis on the input news comment text through a pre-established semantic map and sentiment feature library to generate preliminary compliance assessment results.

[0095] The text optimization processing module 302 is used to optimize the news comment text according to the preliminary compliance assessment results to generate an optimized text that meets the compliance standards.

[0096] The text embedding and distribution module 303 is used to embed the optimized text into the dissemination process through the intelligent streaming media architecture and perform multi-terminal linkage distribution operations.

[0097] The user feedback data acquisition module 304 is used to acquire user feedback data based on the result of the multi-terminal linkage distribution operation.

[0098] The user interaction behavior analysis module 305 is used to perform user interaction behavior analysis on the user feedback data and generate a user acceptance judgment result.

[0099] The virtual scene mapping module 306 is used to map user interaction behaviors to virtual scenes through digital twin technology to generate user experience data.

[0100] The emotional response analysis module 307 is used to analyze the user's emotional response according to the user experience data to obtain an analysis result.

[0101] The recommendation optimization module 308 is used to adjust content recommendation parameters and optimize push logic according to the analysis results, and generate an all-media communication matrix configuration plan.

[0102] The communication effectiveness analysis module 309 is used to analyze the operating data of the full-media communication matrix configuration plan according to preset effectiveness indicators to determine the communication effectiveness and user participation.

[0103] The database optimization module 310 is used to optimize the semantic graph and sentiment feature library based on communication effectiveness and user participation.

[0104] Optional semantic parsing and sentiment analysis modules include:

[0105] Semantic feature extraction unit, used to extract core semantic features from the semantic graph;

[0106] The sentiment tendency analysis unit is used to perform sentiment tendency analysis on news comment text based on the sentiment feature library to obtain the sentiment tendency.

[0107] The dual parsing unit is used to perform dual analysis of the semantic content and emotional expression of news commentary text based on core semantic features and emotional tendencies, and determine the degree of semantic deviation and emotional tendency value.

[0108] The compliance judgment unit is used to judge that the news comment text does not meet the compliance standards when the degree of semantic deviation or the emotional tendency value exceeds a preset threshold.

[0109] The compliance judgment unit is used to comprehensively evaluate the degree of semantic deviation and sentiment tendency value through a multimodal large language model to generate preliminary compliance assessment results.

[0110] Optionally, the text optimization processing module includes:

[0111] The semantic reconstruction unit is used to perform semantic reconstruction on the news comment text when the preliminary compliance assessment results show that there is a semantic deviation or the emotional expression does not meet the preset threshold.

[0112] The emotional expression adjustment unit is used to adjust the emotional expression of the news commentary text and generate an adjusted text that meets the preset emotional characteristics.

[0113] The compliance review unit is used to conduct a compliance review on the adjusted text after semantic reconstruction and sentiment adjustment to obtain a review result.

[0114] The optimization iteration unit is used to repeatedly perform semantic reconstruction and sentiment adjustment operations if the review result does not meet the standard, until the review result meets the standard and an optimized text is obtained.

[0115] Optionally, the virtual scene mapping module includes:

[0116] The behavior model construction unit is used to build a user behavior model based on the interactive behavior in the user feedback data using digital twin technology.

[0117] The virtual scene mapping and rendering unit is used to map the user behavior model into the virtual scene and simulate the immersive news narrative space through the real-time rendering engine.

[0118] The data collection unit is used to obtain the user's operation data and reaction data in the virtual and real environment in the immersive news narrative space.

[0119] The data integration and analysis unit is used to generate user experience data based on the operation data and response data.

[0120] Optional, recommended optimization modules include:

[0121] The emotional response analysis unit is used to calculate the adjustment amount of the content recommendation parameter according to the emotional response characteristics in the analysis result to obtain the adjusted content recommendation parameter.

[0122] The push logic optimization unit is used to dynamically update the timeliness weight and topic relevance coefficient in the push logic based on the preset push logic optimization rules to obtain the updated push logic.

[0123] The configuration scheme generating unit is used to combine and match the adjusted content recommendation parameters with the updated push logic to generate a candidate configuration scheme set.

[0124] The communication effect prediction and scheme selection unit is used to predict the communication effect of the candidate configuration scheme set through Monte Carlo simulation, and select the configuration scheme with the best prediction value as the full media communication matrix configuration scheme.

[0125] An embodiment of the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0126] An embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0127] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0128] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A media service intelligent recommendation method, characterized in that: The method comprises: S1. Perform semantic parsing and sentiment analysis on the input news commentary text using a pre-established semantic graph and sentiment feature library to generate preliminary compliance assessment results. S2. Optimizing the news commentary text based on the preliminary compliance assessment results to generate an optimized text that meets compliance standards; S3. Embed the optimized text into the dissemination process through the intelligent streaming media architecture and perform multi-terminal linkage distribution operations; S4. Obtaining user feedback data based on the result of the multi-terminal linkage distribution operation; S5. Analyze the user interaction behavior of the user feedback data to generate a user acceptance judgment result; S6. Mapping the user interaction behavior to a virtual scene through digital twin technology to generate user experience data; S7. Analyze user emotional responses based on the user experience data to obtain analysis results; adjust content recommendation parameters and optimize push logic based on the analysis results to generate an omnimedia communication matrix configuration plan; S8. According to the preset performance indicators, the operating data of the full-media communication matrix configuration plan is analyzed to determine the communication effectiveness and user participation; based on the communication effectiveness and the user participation, the semantic map and the emotional feature library are optimized.

2. The method according to claim 1, characterized in that Said S1 comprises: S11, extracting core semantic features from the semantic map, performing sentiment analysis on the news comment text based on the sentiment feature library, and obtaining sentiment tendency; S12, performing dual analysis of semantic content and emotional expression on the news commentary text based on the core semantic features and the emotional tendency, and determining a degree of semantic deviation and an emotional tendency value; S13. If the semantic deviation degree or sentiment tendency value exceeds a preset threshold, the news commentary text is determined to be non-compliant with the compliance standard; S14. Comprehensively evaluate the semantic deviation degree and sentiment tendency value through a multimodal large language model to generate the preliminary compliance assessment result.

3. The method according to claim 2, characterized in that The S2 includes: S21. When the preliminary compliance assessment result shows that there is a semantic deviation or the emotional expression does not meet the preset threshold, semantically reconstruct the news comment text; S22, adjusting the emotional expression of the news commentary text to generate an adjusted text that meets preset emotional characteristics; S23, conducting a compliance review on the adjusted text after semantic reconstruction and sentiment adjustment to obtain a review result; S24. If the review result does not meet the standard, the semantic reconstruction and sentiment adjustment operations are repeatedly performed until the review result meets the standard, thereby obtaining the optimized text.

4. The method according to claim 3, characterized in that The S6 includes: S61. Build a user behavior model using digital twin technology based on the interactive behaviors in the user feedback data; S62, mapping the user behavior model to a virtual scene, and simulating an immersive news narrative space through a real-time rendering engine; S63. Acquiring, in the immersive news narrative space, operation data and response data of the user in the virtual-real environment; S64: Generate the user experience data according to the operation data and the response data.

5. The method according to any one of claims 1 to 4, characterized in that The step of adjusting content recommendation parameters and optimizing push logic according to the analysis results to generate an all-media communication matrix configuration plan includes: S71. Calculating an adjustment amount for a content recommendation parameter based on the emotional response characteristics in the analysis result to obtain an adjusted content recommendation parameter. S72. Based on a preset push logic optimization rule, dynamically update the timeliness weight and topic relevance coefficient in the push logic to obtain an updated push logic; S73: Combine and match the adjusted content recommendation parameters with the updated push logic to generate a candidate configuration solution set; S74. Predict the propagation effect of the candidate configuration scheme set through Monte Carlo simulation, and select the configuration scheme with the best prediction value as the all-media propagation matrix configuration scheme.

6. A media service intelligent recommendation device, characterized in that: The device comprises: The semantic parsing and sentiment analysis module is used to perform semantic parsing and sentiment analysis on the input news commentary text using a pre-established semantic graph and sentiment feature library, generating preliminary compliance assessment results. A text optimization processing module, configured to optimize the news comment text according to the preliminary compliance assessment result to generate an optimized text that meets the compliance standards; A text embedding and distribution module is used to embed the optimized text into the dissemination process through an intelligent streaming media architecture and perform multi-terminal linkage distribution operations; A user feedback data acquisition module, configured to acquire user feedback data based on the result of the multi-terminal linkage distribution operation; A user interaction behavior analysis module is used to perform user interaction behavior analysis on the user feedback data and generate a user acceptance judgment result; A virtual scene mapping module is used to map the user interaction behavior to a virtual scene through digital twin technology to generate user experience data; An emotional response analysis module is used to analyze user emotional responses based on the user experience data to obtain analysis results; A recommendation optimization module is used to adjust content recommendation parameters and optimize push logic based on the analysis results, and generate an all-media communication matrix configuration plan; A communication effectiveness analysis module is used to analyze the operating data of the omnimedia communication matrix configuration plan according to preset effectiveness indicators to determine the communication effectiveness and user participation; A database optimization module is used to optimize the semantic map and the emotional feature library based on the communication effectiveness and the user participation.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images