A news intelligent broadcasting method and system based on artificial intelligence

Abstract

The invention discloses a method and system for intelligent news broadcasting based on artificial intelligence, comprising the following steps: S1, using a news crawling tool to obtain data and generate structured event data; S2, using an improved BERT model to generate an event knowledge graph, and using an improved K-Medoids clustering algorithm for cluster analysis; S3, generating a dynamic broadcast script through a multimodal generative adversarial network; S4, analyzing the user's emotional state, and generating personalized voice broadcast content; S5, combining an event-driven mechanism and time-sensitivity optimization rules to generate an enhanced news summary; S6, using a hierarchical attention mechanism to allocate dynamic weights, and combining time-sensitivity optimization rules to dynamically adjust the order of news broadcasting; S7, performing logical consistency detection on news broadcast content; S8, adjusting the broadcast content and generating updated news broadcast content in real time. The present invention combines improved algorithms, sentiment analysis, and dynamic broadcasting technology to achieve personalized news broadcasting.

Description

Technical Field

[0001] The present invention relates to the technical field of news broadcasting, and in particular to an artificial intelligence-based news intelligent broadcasting method and system. Background Art

[0002] With the rapid development of Internet technology, the way of news dissemination has gradually undergone profound changes. The traditional way of news release and dissemination has gradually developed towards intelligence and personalization. The process of news content production, dissemination, acquisition and consumption no longer relies solely on human intervention, but also begins to rely heavily on the assistance of intelligent technology. Against this background, news intelligent broadcasting technology based on artificial intelligence has emerged and has become one of the important development directions in the field of news dissemination.

[0003] Existing news broadcasting technologies are mainly based on traditional automated broadcasting systems. These systems generally rely on simple rule engines and templated text generation methods to process and broadcast news content through predetermined rules. The advantage of these systems is that they realize the automation of news broadcasting and save a lot of manpower costs, but they have obvious limitations in processing capabilities and personalized needs. First, traditional systems often rely only on the basic semantic information of news content, and lack in-depth understanding of the context and semantic relationship between news events, resulting in the generated broadcast content being often relatively simple and unable to provide users with an emotional and personalized experience. Secondly, the existing system cannot accurately perceive and feedback the user's emotions and needs. The broadcast content lacks interactivity and customization with the user's emotional state, interests and hobbies and other personalized factors, resulting in a relatively simple and monotonous user experience.

[0004] In order to improve the intelligence level and personalized service capabilities of news broadcast systems, news broadcast systems based on artificial intelligence technologies such as natural language processing and deep learning have gradually emerged in recent years. In particular, the application of deep learning in text generation, sentiment analysis, speech synthesis, etc. has greatly promoted the development of news broadcast systems. However, there are still many problems with current technologies. First, the existing news broadcast methods mainly rely on a single text or voice model to process news information, lacking the comprehensive use of multimodal data. Although text, voice and image are important components of news communication, existing technologies are often limited to single-modal information processing, making it difficult to achieve multimodal fusion and dynamic broadcast generation. Secondly, although existing sentiment analysis technology can perform emotion recognition on users' voices or facial expressions, it is still insufficient in the recognition and analysis of complex emotional states, and it is difficult to fully understand the subtle emotional changes of users. In addition, the timeliness and semantic relationships between news events often lack effective modeling and reasoning in existing systems, making the news broadcast content lack logic and consistency, and users cannot obtain dynamically adjusted broadcast content that meets actual needs.

[0005] Although the current intelligent news broadcasting technology has made some breakthroughs, there are still many areas for improvement in the existing technology. For example, in the generation of event knowledge graphs and semantic feature analysis, although deep learning and clustering analysis can effectively identify and process the semantic relationships of news events, how to further improve the accuracy of graph construction, reduce clustering errors, and how to further optimize the selection and representation of semantic features are still topics worthy of in-depth research. In addition, the accuracy and real-time performance of sentiment analysis models are also issues that need to be urgently addressed by current technology, especially when dynamically adjusting the content of news broadcasts. How to combine users' real-time emotional changes, personal interests and needs to accurately adjust the tone, speed and emotional color of the broadcast content still poses great technical challenges.

[0006] Therefore, how to provide a news intelligent broadcasting method and system based on artificial intelligence is a problem that technical personnel in this field need to solve urgently. Summary of the invention

[0007] One purpose of the present invention is to propose a method and system for intelligent news broadcasting based on artificial intelligence. The present invention combines an improved BERT model to generate an event knowledge graph, and uses an improved K-Medoids clustering algorithm to optimize and analyze event semantic features. At the same time, it uses multimodal generative adversarial networks, sentiment analysis, and dynamic broadcasting technology to achieve personalized news broadcasting based on the user's emotional state. Through multi-level data processing and semantic optimization, the method can automatically extract key events, generate logically consistent news summaries, and dynamically adjust the voice broadcast content according to the user's real-time emotions. It has the advantages of high efficiency, accuracy, and strong user emotional adaptability, which significantly improves the intelligence of news dissemination and user experience.

[0008] According to an embodiment of the present invention, a news intelligent broadcasting method based on artificial intelligence includes the following steps:

[0009] S1. Use news crawling tools to obtain news text data, news image data, and news video data, and extract and parse them to generate structured event data;

[0010] S2. Based on structured event data, the improved BERT model is used to generate event knowledge graphs, and the improved K-Medoids clustering algorithm is used to cluster the semantic features of events and generate priority tags;

[0011] S3, based on the semantic feature data and priority tags in the event knowledge graph, a dynamic broadcast script is generated through a multimodal generative adversarial network;

[0012] S4, combining the dynamic broadcast script, the user's voice signal data and the user's facial expression data, using the emotion detection model to analyze the user's emotional state, and generating personalized voice broadcast content that matches the user's emotional state through the emotion-driven speech synthesis model;

[0013] S5. Based on the event knowledge graph, combined with the event-driven mechanism and time-sensitive optimization rules, summary generation is performed to generate enhanced news summaries;

[0014] S6. Based on the title, text and key sentence data in the enhanced news summary, the hierarchical attention mechanism is used to assign dynamic weights, and the news broadcast order is dynamically adjusted in combination with the time-sensitive optimization rule;

[0015] S7. Perform logical consistency detection on the dynamically adjusted news broadcast content, and automatically resolve the contradictory content based on the semantic association and credibility score of the event knowledge graph to generate logically consistent optimized broadcast content;

[0016] S8. Based on the logically consistent optimized broadcast content, combined with the user's real-time voice command data or the user's gesture signal data, the optimized broadcast content is adjusted, and updated news broadcast content is generated in real time.

[0017] Optionally, the S2 specifically includes:

[0018] S21. Based on structured event data, an improved BERT model is used to semantically encode events to obtain a set of semantic embedding vectors corresponding to the event set. The improvement of the BERT model introduces an event difference kernel function in the self-attention mechanism. , used to characterize the differences in spatiotemporal characteristics between events:

[0019] ;

[0020] in, represents the output weight matrix of the improved self-attention mechanism, Q represents the query matrix in the self-attention mechanism inside BERT, K represents the key matrix in the self-attention mechanism inside BERT, and V represents the value matrix in the self-attention mechanism inside BERT. represents element-wise multiplication, represents the scaling factor, Indicates converting the input value into a probability distribution;

[0021] ;

[0022] in, Indicates an event and events The spatial-temporal difference kernel function value between the two, exp represents the exponential function, Indicates an event and events The difference function output value in the time feature space, Indicates an event and events The difference function output value in the location feature space, , and Indicates adjustment parameters;

[0023] S22. Construct an event knowledge graph based on the semantic embedding vector set, take the event subject, time and location as graph nodes, and calculate the edge weights between graph nodes:

[0024] ;

[0025] in, Indicates an event and events The edge weights between nodes, Indicates an event and events The cosine similarity value between the semantic embedding vectors of ;

[0026] S23. Improved K-Medoids clustering analysis of the semantic embedding vector set in the event knowledge graph. The improvements include introducing a polynomial kernel function to impose nonlinear constraints on cluster center selection and update, and dynamically balancing the semantic distance and event difference kernel in cluster iteration:

[0027] ;

[0028] in, represents the event nodes to be clustered, represents the jth cluster center node, represents the clustering cost, represents the Euclidean distance, , and represents the learnable parameters, Represents the event node to be clustered The semantic embedding vector of Represents the jth cluster center node The semantic embedding vector of

[0029] S24. Calculate the relevance score for the event nodes to be clustered based on the clustering results:

[0030] ;

[0031] in, represents the relevance score, represents the event density, , and represents the weight coefficient;

[0032] S25. Calculate the priority score by combining the relevance score with the importance characteristics of the event:

[0033] ;

[0034] in, represents the priority score, Indicates the importance of the event. and represents the weight coefficient;

[0035] S26. Generate priority tags for nodes in the event knowledge graph based on the priority scores, and store the priority tags in the event knowledge graph.

[0036] Optionally, the S3 specifically includes:

[0037] S31, extracting semantic feature data related to news events from the event knowledge graph, and combining the priority tags of the events to construct a multimodal input data set;

[0038] S32, inputting the multimodal input data set into a generator model of a multimodal generative adversarial network, wherein the generator model includes a text processing module and a visual information processing module to generate text features and visual features;

[0039] S33, in the generator model, deeply integrating text features and visual features through a cross-modal mapping module to generate a preliminary dynamic broadcast script containing text and visual information;

[0040] S34, inputting the generated preliminary dynamic broadcast script into the discriminator model, the discriminator model evaluates the matching degree between the text information and the visual information of the generated broadcast script and the semantic consistency, and the discriminator judges the quality of the preliminary dynamic broadcast script by comparing the similarity between the generated preliminary dynamic broadcast script and the real news data, and outputs optimization feedback;

[0041] S35. Based on adversarial training of the generator model and the discriminator model, the generation strategy of the generator model is adjusted through multiple rounds of feedback to optimize the preliminary dynamic broadcast script;

[0042] S36. Sort the generated preliminary dynamic broadcast scripts in combination with the priority tags, adjust the broadcast order according to the priority of the event and the degree of social attention, and output the final sorted dynamic broadcast script.

[0043] Optionally, the S4 specifically includes:

[0044] S41, combining the dynamic broadcast script, obtaining the user's voice signal data and facial expression data, receiving the user's voice input in real time through the voice acquisition device, and capturing the user's facial expression image through the camera device;

[0045] S42, using an emotion detection model to perform emotion analysis on the user's voice signal data and facial expression data, wherein the emotion detection model includes a voice emotion analysis module and a facial expression recognition module, which process the voice signal and the facial image respectively, and extract the voice emotion features and the facial expression features;

[0046] S43, integrating the voice emotion features with the facial expression features to generate a comprehensive emotion state score:

[0047] ;

[0048] in, represents the comprehensive emotional state score, represents the speech emotion feature score, represents the facial expression feature score, and represents the weight factor, represents the saturation coefficient;

[0049] S44, based on the comprehensive emotional state score, calling the emotion-driven speech synthesis model, the speech synthesis model dynamically adjusts the speech synthesis parameters according to the user's emotional state, including voice, speech speed and emotional color, to generate personalized voice broadcast content that matches the user's emotional state;

[0050] S45. In the emotion-driven speech synthesis process, the speech synthesis parameters are dynamically adjusted according to the real-time emotional changes of the user:

[0051] ;

[0052] in, Indicates the adjusted speech synthesis parameters, represents the initial speech synthesis parameters, and represents the adjustment coefficient, tanh represents the hyperbolic tangent function, and represents the regulating factor;

[0053] S46: Output the speech synthesis result adjusted by emotion-driven as the final personalized speech broadcast content to the user's playback device.

[0054] Optionally, the S5 specifically includes:

[0055] S51, extracting news event data from the event knowledge graph, wherein the news event data includes the subject, time, location, attributes and interrelated information of the event;

[0056] S52, screening news events based on an event-driven mechanism, determining key events that need to be processed first by defining event triggering conditions, and forming an event set to be summarized;

[0057] S53, sorting the event set to be generated by summarization using a time sensitivity optimization rule, wherein the time sensitivity optimization rule adjusts the priority of the event according to the time of occurrence, duration, and impact on the user of the event;

[0058] S54, performing importance scoring on the sorted set of events to be summarized, wherein the importance scoring comprehensively considers the social influence of the event, the public attention, and the degree of association of the event in the knowledge graph;

[0059] S55, screening the event set according to the importance score, selecting several events with the highest scores as core events for summary generation, and generating preliminary content of the news summary;

[0060] S56. Based on the contextual information of the selected core events, natural language generation technology is used for further integration to generate an enhanced news summary containing key information.

[0061] Optionally, the S6 specifically includes:

[0062] S61, assigning initial weights by analyzing the title, text and key sentence data in the enhanced news summary, and adjusting the weights according to the importance, timeliness and relevance of the news content;

[0063] S62, processing the title data, and dynamically adjusting the weight of the title based on the keywords in the title and the relevance in the event knowledge graph;

[0064] S63, further processing the text data, by analyzing the contextual relationship and paragraph importance of the text, and using a hierarchical attention mechanism to assign weights to different paragraphs or sentences in the text;

[0065] S64, processing the key sentence data, and dynamically adjusting the weight of the key sentence according to the position of the sentence in the entire news, the context, and the importance of the explanation of the event background;

[0066] S65. Introduce time-sensitivity optimization rules to dynamically adjust the news broadcast order of each part of the content according to the occurrence time, duration and immediate impact of the event on the user;

[0067] S66. Apply the optimized weights and broadcast order to the final news broadcast script.

[0068] Optionally, the S7 specifically includes:

[0069] S71. Performing logical consistency detection on the dynamically adjusted news broadcast content, using natural language processing technology to analyze the logical structure and semantic association of the news content, and identifying potential logical contradictions or inconsistencies;

[0070] S72. Calculate the semantic similarity between paragraphs based on the semantic association of the event knowledge graph:

[0071] ;

[0072] in, Indicates the overall logical conflict degree, represents the semantic similarity between paragraph m and paragraph n, represents the association strength between event m and event n in the event knowledge graph, M represents the total number of paragraphs, and Q represents the total number of events;

[0073] S73, using a credibility scoring mechanism to evaluate the credibility of the identified logical conflict paragraphs, and deciding whether to automatically resolve them based on the scoring results;

[0074] S74. Combine the semantic association of the event knowledge graph to automatically select the most appropriate alternative content to resolve the conflicting paragraphs:

[0075] ;

[0076] in, Indicates alternative information, represents the set of all candidate nodes related to paragraph m, represents the weight of candidate node j, represents the semantic similarity between candidate node j and paragraph m, and R represents the number of candidate nodes;

[0077] S75. Score the logical consistency of the digested news content, and generate and output logically consistent optimized broadcast content.

[0078] An artificial intelligence-based news intelligent broadcasting system according to an embodiment of the present invention includes the following modules:

[0079] News data acquisition module, used to collect data and generate structured event data;

[0080] The event knowledge graph generation module is used to generate an event knowledge graph based on structured event data, perform cluster analysis, and generate priority tags;

[0081] A dynamic broadcast script generation module is used to generate dynamic broadcast scripts through a multimodal generative adversarial network;

[0082] Emotion analysis module, used to analyze the user's voice signal and facial expression data to obtain the emotional state;

[0083] Personalized speech synthesis module, used to generate personalized speech broadcast content according to user emotions;

[0084] A news summary generation module, used to generate enhanced news summaries;

[0085] The broadcast content sorting and optimization module is used to adjust the order of news broadcasts and assign weights;

[0086] Logical consistency detection module, used to detect and resolve logical contradictions and generate optimized broadcast content;

[0087] The news broadcast content adjustment module is used to adjust and update the news broadcast content in real time according to the user's real-time voice command data or the user's gesture signal data.

[0088] The beneficial effects of the present invention are:

[0089] First, the event knowledge graph generated by the improved BERT model and the K-Medoids clustering algorithm can effectively analyze and model the semantic features of news events. This method can not only deeply explore the spatiotemporal correlation between events, but also identify the multi-dimensional characteristics of news content, such as social influence, timeliness, and public attention, so that the generation of news broadcast content is more in line with the actual importance and social significance of the event. In addition, based on these optimized semantic analysis results, the generated dynamic broadcast script can better integrate multimodal information (such as text, images, videos, etc.), providing users with richer and more diverse broadcast content.

[0090] Secondly, the present invention makes full use of sentiment analysis technology and personalized speech synthesis models to capture and analyze the user's emotional state in real time. Through emotion detection of the user's voice signal and facial expression data, the system can generate personalized voice broadcast content that matches the user's emotions. This feature not only enhances the emotional color and interactivity of news broadcasts, but also dynamically adjusts the tone, speed and emotional expression of the broadcast according to the user's emotional fluctuations, making the broadcast content more in line with the user's psychological needs and enhancing the user's sense of immersion and personalized experience.

[0091] In addition, the present invention also generates and sorts news summaries through event-driven mechanisms and time-sensitive optimization rules, thereby ensuring that the broadcast content is more accurate and timely. By scoring the importance of events and adjusting the timeliness, the news summaries can dynamically adjust their broadcast order according to social attention, the continuity of events, and the immediate impact on users. This optimization mechanism makes the broadcast content not only a simple retelling of the news, but also an intelligent output that is updated and adjusted in real time according to the time and space background and user needs.

[0092] In terms of the logical consistency of news broadcast content, the present invention can effectively detect and resolve possible logical contradictions when dynamically adjusting the broadcast order by combining the semantic association of event knowledge graphs. The system can automatically identify and correct inconsistencies in broadcasts based on in-depth analysis of the semantics of news content, ensuring that the final output broadcast content is highly logical and credible.

[0093] Finally, the system design of the present invention is highly flexible and extensible, and can further adjust and optimize the content according to the user's real-time feedback (such as voice commands and gesture signals) during the news broadcast process. Through this adaptive mechanism, the system can adjust the broadcast content in real time according to the changes in user needs, enhance the interactivity and personalization of news broadcasts, and further improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0095] Figure 1 This is an overall flow chart of an artificial intelligence-based news intelligent broadcasting method proposed by the present invention;

[0096] Figure 2 This is a structural schematic diagram of a multimodal generative adversarial network for generating dynamic broadcast scripts for an artificial intelligence-based news intelligent broadcast system proposed by the present invention. DETAILED DESCRIPTION

[0097] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0098] refer to Figure 1 , a news intelligent broadcasting method based on artificial intelligence, comprising the following steps:

[0099] S1. Use news crawling tools to obtain news text data, news image data, and news video data, and extract and parse them to generate structured event data;

[0100] S2. Based on structured event data, the improved BERT model is used to generate event knowledge graphs, and the improved K-Medoids clustering algorithm is used to cluster the semantic features of events and generate priority tags;

[0101] S3, based on the semantic feature data and priority tags in the event knowledge graph, a dynamic broadcast script is generated through a multimodal generative adversarial network;

[0102] S4, combining the dynamic broadcast script, the user's voice signal data and the user's facial expression data, using the emotion detection model to analyze the user's emotional state, and generating personalized voice broadcast content that matches the user's emotional state through the emotion-driven speech synthesis model;

[0103] S5. Based on the event knowledge graph, combined with the event-driven mechanism and time-sensitive optimization rules, summary generation is performed to generate enhanced news summaries;

[0104] S6. Based on the title, text and key sentence data in the enhanced news summary, the hierarchical attention mechanism is used to assign dynamic weights, and the news broadcast order is dynamically adjusted in combination with the time-sensitive optimization rule;

[0105] S7. Perform logical consistency detection on the dynamically adjusted news broadcast content, and automatically resolve the contradictory content based on the semantic association and credibility score of the event knowledge graph to generate logically consistent optimized broadcast content;

[0106] S8. Based on the logically consistent optimized broadcast content, combined with the user's real-time voice command data or the user's gesture signal data, the optimized broadcast content is adjusted, and updated news broadcast content is generated in real time.

[0107] In this implementation, S2 specifically includes:

[0108] S21. Based on structured event data, an improved BERT model is used to semantically encode events to obtain a set of semantic embedding vectors corresponding to the event set. The improvement of the BERT model introduces an event difference kernel function in the self-attention mechanism. , used to characterize the differences in spatiotemporal characteristics between events:

[0109] ;

[0110] in, represents the output weight matrix of the improved self-attention mechanism, Q represents the query matrix in the self-attention mechanism inside BERT, K represents the key matrix in the self-attention mechanism inside BERT, and V represents the value matrix in the self-attention mechanism inside BERT. represents element-wise multiplication, represents the scaling factor, Indicates converting the input value into a probability distribution;

[0111] ;

[0112] in, Indicates an event and events The spatial-temporal difference kernel function value between the two, exp represents the exponential function, Indicates an event and events The difference function output value in the time feature space, Indicates an event and events The difference function output value in the location feature space, , and Indicates adjustment parameters;

[0113] S22. Construct an event knowledge graph based on the semantic embedding vector set, take the event subject, time and location as graph nodes, and calculate the edge weights between graph nodes:

[0114] ;

[0115] in, Indicates an event and events The edge weights between nodes, Indicates an event and events The cosine similarity value between the semantic embedding vectors of ;

[0116] S23. Improved K-Medoids clustering analysis of the semantic embedding vector set in the event knowledge graph. The improvements include introducing a polynomial kernel function to impose nonlinear constraints on cluster center selection and update, and dynamically balancing the semantic distance and event difference kernel in cluster iteration:

[0117] ;

[0118] in, represents the event nodes to be clustered, represents the jth cluster center node, represents the clustering cost, represents the Euclidean distance, , and represents the learnable parameters, Represents the event node to be clustered The semantic embedding vector of Represents the jth cluster center node The semantic embedding vector of

[0119] S24. Calculate the relevance score for the event nodes to be clustered based on the clustering results:

[0120] ;

[0121] in, represents the relevance score, represents the event density, , and represents the weight coefficient;

[0122] S25. Calculate the priority score by combining the relevance score with the importance characteristics of the event:

[0123] ;

[0124] in, represents the priority score, Indicates the importance of the event. and represents the weight coefficient;

[0125] S26. Generate priority tags for nodes in the event knowledge graph based on the priority scores, and store the priority tags in the event knowledge graph.

[0126] In this implementation, S3 specifically includes:

[0127] S31, extracting semantic feature data related to news events from the event knowledge graph, and combining the priority tags of the events to construct a multimodal input data set;

[0128] S32, inputting the multimodal input data set into a generator model of a multimodal generative adversarial network, wherein the generator model includes a text processing module and a visual information processing module to generate text features and visual features;

[0129] S33, in the generator model, deeply integrating text features and visual features through a cross-modal mapping module to generate a preliminary dynamic broadcast script containing text and visual information;

[0130] S34, inputting the generated preliminary dynamic broadcast script into the discriminator model, the discriminator model evaluates the matching degree between the text information and the visual information of the generated broadcast script and the semantic consistency, and the discriminator judges the quality of the preliminary dynamic broadcast script by comparing the similarity between the generated preliminary dynamic broadcast script and the real news data, and outputs optimization feedback;

[0131] S35. Based on adversarial training of the generator model and the discriminator model, the generation strategy of the generator model is adjusted through multiple rounds of feedback to optimize the preliminary dynamic broadcast script;

[0132] S36. Sort the generated preliminary dynamic broadcast scripts in combination with the priority tags, adjust the broadcast order according to the priority of the event and the degree of social attention, and output the final sorted dynamic broadcast script.

[0133] In this implementation manner, the S4 specifically includes:

[0134] S41, combining the dynamic broadcast script, obtaining the user's voice signal data and facial expression data, receiving the user's voice input in real time through the voice acquisition device, and capturing the user's facial expression image through the camera device;

[0135] S42, using an emotion detection model to perform emotion analysis on the user's voice signal data and facial expression data, wherein the emotion detection model includes a voice emotion analysis module and a facial expression recognition module, which process the voice signal and the facial image respectively, and extract the voice emotion features and the facial expression features;

[0136] S43, integrating the voice emotion features with the facial expression features to generate a comprehensive emotion state score:

[0137] ;

[0138] in, represents the comprehensive emotional state score, represents the speech emotion feature score, represents the facial expression feature score, and represents the weight factor, represents the saturation coefficient;

[0139] S44, based on the comprehensive emotional state score, calling the emotion-driven speech synthesis model, the speech synthesis model dynamically adjusts the speech synthesis parameters according to the user's emotional state, including voice, speech speed and emotional color, to generate personalized voice broadcast content that matches the user's emotional state;

[0140] S45. In the emotion-driven speech synthesis process, the speech synthesis parameters are dynamically adjusted according to the real-time emotional changes of the user:

[0141] ;

[0142] in, Indicates the adjusted speech synthesis parameters, represents the initial speech synthesis parameters, and represents the adjustment coefficient, tanh represents the hyperbolic tangent function, and represents the regulating factor;

[0143] S46: Output the speech synthesis result adjusted by emotion-driven as the final personalized speech broadcast content to the user's playback device.

[0144] In this implementation manner, S5 specifically includes:

[0145] S51, extracting news event data from the event knowledge graph, wherein the news event data includes the subject, time, location, attributes and interrelated information of the event;

[0146] S52, screening news events based on an event-driven mechanism, determining key events that need to be processed first by defining event triggering conditions, and forming an event set to be summarized;

[0147] S53, sorting the event set to be generated by summarization using a time sensitivity optimization rule, wherein the time sensitivity optimization rule adjusts the priority of the event according to the time of occurrence, duration, and impact on the user of the event;

[0148] S54, performing importance scoring on the sorted set of events to be summarized, wherein the importance scoring comprehensively considers the social influence of the event, the public attention, and the degree of association of the event in the knowledge graph;

[0149] S55, screening the event set according to the importance score, selecting several events with the highest scores as core events for summary generation, and generating preliminary content of the news summary;

[0150] S56. Based on the contextual information of the selected core events, natural language generation technology is used for further integration to generate an enhanced news summary containing key information.

[0151] In this implementation manner, S6 specifically includes:

[0152] S61, assigning initial weights by analyzing the title, text and key sentence data in the enhanced news summary, and adjusting the weights according to the importance, timeliness and relevance of the news content;

[0153] S62, processing the title data, and dynamically adjusting the weight of the title based on the keywords in the title and the relevance in the event knowledge graph;

[0154] S63, further processing the text data, by analyzing the contextual relationship and paragraph importance of the text, and using a hierarchical attention mechanism to assign weights to different paragraphs or sentences in the text;

[0155] S64, processing the key sentence data, and dynamically adjusting the weight of the key sentence according to the position of the sentence in the entire news, the context, and the importance of the explanation of the event background;

[0156] S65. Introduce time-sensitivity optimization rules to dynamically adjust the news broadcast order of each part of the content according to the occurrence time, duration and immediate impact of the event on the user;

[0157] S66. Apply the optimized weights and broadcast order to the final news broadcast script.

[0158] In this implementation manner, the S7 specifically includes:

[0159] S71. Performing logical consistency detection on the dynamically adjusted news broadcast content, using natural language processing technology to analyze the logical structure and semantic association of the news content, and identifying potential logical contradictions or inconsistencies;

[0160] S72. Calculate the semantic similarity between paragraphs based on the semantic association of the event knowledge graph:

[0161] ;

[0162] in, Indicates the overall logical conflict degree, represents the semantic similarity between paragraph m and paragraph n, represents the association strength between event m and event n in the event knowledge graph, M represents the total number of paragraphs, and Q represents the total number of events;

[0163] S73, using a credibility scoring mechanism to evaluate the credibility of the identified logical conflict paragraphs, and deciding whether to automatically resolve them based on the scoring results;

[0164] S74. Combine the semantic association of the event knowledge graph to automatically select the most appropriate alternative content to resolve the conflicting paragraphs:

[0165] ;

[0166] in, Indicates alternative information, represents the set of all candidate nodes related to paragraph m, represents the weight of candidate node j, represents the semantic similarity between candidate node j and paragraph m, and R represents the number of candidate nodes;

[0167] S75. Score the logical consistency of the digested news content, and generate and output logically consistent optimized broadcast content.

[0168] refer to Figure 2 , an artificial intelligence-based news intelligent broadcasting system, including the following modules:

[0169] News data acquisition module, used to collect data and generate structured event data;

[0170] The event knowledge graph generation module is used to generate an event knowledge graph based on structured event data, perform cluster analysis, and generate priority tags;

[0171] A dynamic broadcast script generation module is used to generate dynamic broadcast scripts through a multimodal generative adversarial network;

[0172] Emotion analysis module, used to analyze the user's voice signal and facial expression data to obtain the emotional state;

[0173] Personalized speech synthesis module, used to generate personalized speech broadcast content according to user emotions;

[0174] A news summary generation module, used to generate enhanced news summaries;

[0175] The broadcast content sorting and optimization module is used to adjust the order of news broadcasts and assign weights;

[0176] Logical consistency detection module, used to detect and resolve logical contradictions and generate optimized broadcast content;

[0177] The news broadcast content adjustment module is used to adjust and update the news broadcast content in real time according to the user's real-time voice command data or the user's gesture signal data.

[0178] Embodiment 1:

[0179] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the intelligent news broadcasting scenario of a sudden typhoon event in a coastal city. The intensity of the typhoon reached a super-strong level. In this application, the present invention first crawls and parses news text, image and video data from multiple channels (including government emergency announcements, meteorological department reports, mainstream media reports and real-time dynamics of social media) through the news data acquisition module. These data are converted into structured event data, such as event subject (typhoon and its path), time (specific occurrence time and predicted duration), location (affected area) and event impact (number of evacuees, economic losses, etc.). The system uses the improved BERT model to generate an event knowledge graph, and constructs a dynamic and real-time knowledge network by capturing changes in the typhoon path, expansion of the affected area and updates of emergency measures.

[0180] Based on the event knowledge graph, the system analyzes the semantic features of events through an improved K-Medoids clustering algorithm. The focus is on typhoon path changes, rescue responses, and high-frequency topics of public concern, such as traffic disruption information and conditions in disaster-stricken communities. Each event node is marked with a priority to ensure that the information most relevant to public needs is broadcast first. For example, real-time changes in the typhoon path and evacuation recommendations are set to the highest priority, while non-urgent but still relevant background data is set to a lower priority.

[0181] Subsequently, based on the priority marking, the system uses a multimodal generative adversarial network to generate a dynamic broadcast script. The multimodal generative adversarial network integrates text data, image data, and video data, and generates multimodal broadcast content including typhoon path dynamic map, real-time pictures of the disaster-stricken area, and key text information through cross-modal mapping technology. The generated preliminary dynamic broadcast script is evaluated for matching and logical consistency through the discriminator model to ensure the authenticity and consistency of the content. The optimized broadcast script is sorted according to the priority and social attention of the event to generate the final dynamic broadcast content that best suits the public's needs.

[0182] During the news broadcast process, the present invention further combines the user's voice signal and facial expression data to analyze the public's emotional state through an emotion detection model. The results of the emotion analysis show that the public mainly expressed tension and worry in the early stages of the disaster, and gradually transitioned to hope and confidence as the rescue operations gradually progressed. The present invention uses an emotion-driven speech synthesis model to adjust the tone, speed and emotional color of the broadcast according to real-time emotions, so that the broadcast content is more in line with the public's emotional needs. For example, when issuing evacuation recommendations, the voice broadcast uses a serious and fast tone, while when conveying successful rescue cases, a gentle and confident tone is used.

[0183] In order to evaluate the actual effect of the present invention, we selected 1,000 public users and conducted a follow-up survey on the performance of the system during the typhoon event. The main evaluation indicators include the accuracy of news information transmission, the fit between the broadcast content and public sentiment, and the user's satisfaction with the broadcast service. The experimental results show that the intelligent broadcast system of the present invention significantly improves the accuracy, timeliness and interactivity of news transmission.

[0184] Table 1 Comparison of the accuracy of information transmission between the system of the present invention and the traditional system

[0185]

[0186] Table 2 Comparison of user satisfaction scores between the system of the present invention and the traditional system

[0187]

[0188] It can be seen from the results in Table 1 above that the system of the present invention shows significant advantages in the accuracy of news information transmission. Within the first two hours of the emergency, the system's transmission accuracy reached 95.6%, which is much higher than the 82.3% of the traditional system. As the event develops, the accuracy of the system of the present invention is further improved to 96.8% within 2 to 4 hours, while the traditional system is only 85.7%. Within 4 to 8 hours of the event, the system performs even better, with an accuracy of 98.2%, an increase of 12.3% over the 87.5% of the traditional system. From the overall average, the transmission accuracy of the system of the present invention is 96.9%, significantly higher than the 85.2% of the traditional system. These data fully prove that the present invention can accurately extract key event information and generate logically consistent broadcast content through the improved BERT model to generate event knowledge graphs and the optimized K-Medoids clustering algorithm, thereby efficiently realizing news dissemination in time-sensitive emergencies.

[0189] From the perspective of user satisfaction, the above Table 2 shows that the system of the present invention has been highly recognized by users in multiple dimensions. In terms of the accuracy of the broadcast content, the user score was 4.9 points, significantly higher than the 4.1 points of the traditional system; in terms of broadcast interactivity, the system of the present invention scored 4.7 points, while the traditional system scored 3.8 points. The final user's overall satisfaction score for the system of the present invention reached 4.8 points, while the traditional system was only 4.0 points. These data show that the present invention has significant advantages in meeting the personalized needs of users and improving user experience. Through real-time dynamic adjustment of broadcast content and efficient news content generation, the present invention provides users with more accurate, intelligent and interactive news broadcast services.

[0190] Comprehensive analysis shows that the present invention has shown comprehensive superiority in the intelligent news broadcasting of emergencies. It far exceeds traditional methods in terms of the accuracy of news information transmission, emotional adaptability and user satisfaction. It not only improves the efficiency and quality of news dissemination, but also significantly optimizes the user experience through improved deep learning models and multimodal generation technology, combined with sentiment analysis and dynamic adjustment of broadcasting mechanisms. These results prove that the present invention provides an effective solution for the field of intelligent news broadcasting, which has broad application value and practical significance.

[0191] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A news intelligent broadcasting method based on artificial intelligence, characterized in that: The steps include: S1. Use news crawling tools to obtain news text data, news image data, and news video data, and extract and parse them to generate structured event data; S2. Based on structured event data, the improved BERT model is used to generate event knowledge graphs, and the improved K-Medoids clustering algorithm is used to cluster the semantic features of events and generate priority tags; The improvement of the BERT model introduces the event difference kernel function in the self-attention mechanism , used to characterize the differences in spatiotemporal characteristics between events: ； in, represents the output weight matrix of the improved self-attention mechanism, Q represents the query matrix in the self-attention mechanism inside BERT, K represents the key matrix in the self-attention mechanism inside BERT, and V represents the value matrix in the self-attention mechanism inside BERT. represents element-wise multiplication, represents the scaling factor, Indicates converting the input value into a probability distribution; ； in, Indicates an event and events The spatial-temporal difference kernel function value between the two, exp represents the exponential function, Indicates an event and events The difference function output value in the time feature space, Indicates an event and events The difference function output value in the location feature space, , and Indicates adjustment parameters; The improvements of the improved K-Medoids clustering algorithm include introducing a polynomial kernel function to impose nonlinear constraints on cluster center selection and update, and dynamically balancing the semantic distance and event difference kernel in clustering iterations: ； in, represents the event nodes to be clustered, represents the jth cluster center node, represents the clustering cost, represents the Euclidean distance, , and represents the learnable parameters, Represents the event node to be clustered The semantic embedding vector of Represents the jth cluster center node The semantic embedding vector of S3, based on the semantic feature data and priority tags in the event knowledge graph, a dynamic broadcast script is generated through a multimodal generative adversarial network; S4, combining the dynamic broadcast script, the user's voice signal data and the user's facial expression data, using the emotion detection model to analyze the user's emotional state, and generating personalized voice broadcast content that matches the user's emotional state through the emotion-driven speech synthesis model; S5. Based on the event knowledge graph, combined with the event-driven mechanism and time-sensitive optimization rules, summary generation is performed to generate enhanced news summaries; S6. Based on the title, text and key sentence data in the enhanced news summary, the hierarchical attention mechanism is used to assign dynamic weights, and the news broadcast order is dynamically adjusted in combination with the time-sensitive optimization rule; S7. Perform logical consistency detection on the dynamically adjusted news broadcast content, and automatically resolve the contradictory content based on the semantic association and credibility score of the event knowledge graph to generate logically consistent optimized broadcast content; S8. Based on the logically consistent optimized broadcast content, combined with the user's real-time voice command data or the user's gesture signal data, the optimized broadcast content is adjusted, and updated news broadcast content is generated in real time.

2. The method for intelligent news broadcasting based on artificial intelligence according to claim 1, characterized in that: The S2 specifically includes: S21. Based on the structured event data, the improved BERT model is used to semantically encode the events and obtain the semantic embedding vector set corresponding to the event set; S22. Construct an event knowledge graph based on the semantic embedding vector set, take the event subject, time and location as graph nodes, and calculate the edge weights between graph nodes: ； in, Indicates an event and events The edge weights between nodes, Indicates an event and events The cosine similarity value between the semantic embedding vectors of ; S23, performing improved K-Medoids clustering analysis on the semantic embedding vector set in the event knowledge graph; S24. Calculate the relevance score for the event nodes to be clustered based on the clustering results: ； in, represents the relevance score, represents the event density, , and represents the weight coefficient; S25. Calculate the priority score by combining the relevance score with the importance characteristics of the event: ； in, represents the priority score, Indicates the importance of the event. and represents the weight coefficient; S26. Generate priority tags for nodes in the event knowledge graph based on the priority scores, and store the priority tags in the event knowledge graph.

3. The method for intelligent news broadcasting based on artificial intelligence according to claim 1, characterized in that: The S3 specifically includes: S31, extracting semantic feature data related to news events from the event knowledge graph, and combining the priority tags of the events to construct a multimodal input data set; S32, inputting the multimodal input data set into a generator model of a multimodal generative adversarial network, wherein the generator model includes a text processing module and a visual information processing module to generate text features and visual features; S33, in the generator model, deeply integrating text features and visual features through a cross-modal mapping module to generate a preliminary dynamic broadcast script containing text and visual information; S34, inputting the generated preliminary dynamic broadcast script into the discriminator model, the discriminator model evaluates the matching degree between the text information and the visual information of the generated broadcast script and the semantic consistency, and the discriminator judges the quality of the preliminary dynamic broadcast script by comparing the similarity between the generated preliminary dynamic broadcast script and the real news data, and outputs optimization feedback; S35. Based on adversarial training of the generator model and the discriminator model, the generation strategy of the generator model is adjusted through multiple rounds of feedback to optimize the preliminary dynamic broadcast script; S36. Sort the generated preliminary dynamic broadcast scripts in combination with the priority tags, adjust the broadcast order according to the priority of the event and the degree of social attention, and output the final sorted dynamic broadcast script.

4. The method for intelligent news broadcasting based on artificial intelligence according to claim 1, characterized in that: The S4 specifically includes: S41, combining the dynamic broadcast script, obtaining the user's voice signal data and facial expression data, receiving the user's voice input in real time through the voice acquisition device, and capturing the user's facial expression image through the camera device; S42, using an emotion detection model to perform emotion analysis on the user's voice signal data and facial expression data, wherein the emotion detection model includes a voice emotion analysis module and a facial expression recognition module, which process the voice signal and the facial image respectively, and extract the voice emotion features and the facial expression features; S43, integrating the voice emotion features with the facial expression features to generate a comprehensive emotion state score: ； in, represents the comprehensive emotional state score, represents the speech emotion feature score, represents the facial expression feature score, and represents the weight factor, represents the saturation coefficient; S44, based on the comprehensive emotional state score, calling the emotion-driven speech synthesis model, the speech synthesis model dynamically adjusts the speech synthesis parameters according to the user's emotional state, including voice, speech speed and emotional color, to generate personalized voice broadcast content that matches the user's emotional state; S45. In the emotion-driven speech synthesis process, the speech synthesis parameters are dynamically adjusted according to the real-time emotional changes of the user: ； in, Indicates the adjusted speech synthesis parameters, represents the initial speech synthesis parameters, and represents the adjustment coefficient, tanh represents the hyperbolic tangent function, and represents the regulating factor; S46: Output the speech synthesis result adjusted by emotion-driven as the final personalized speech broadcast content to the user's playback device.

5. The method for intelligent news broadcasting based on artificial intelligence according to claim 1, characterized in that: The S5 specifically includes: S51, extracting news event data from the event knowledge graph, wherein the news event data includes the subject, time, location, attributes and interrelated information of the event; S52, screening news events based on an event-driven mechanism, determining key events that need to be processed first by defining event triggering conditions, and forming an event set to be summarized; S53, sorting the event set to be generated by summarization using a time sensitivity optimization rule, wherein the time sensitivity optimization rule adjusts the priority of the event according to the time of occurrence, duration, and impact on the user of the event; S54, performing importance scoring on the sorted set of events to be summarized, wherein the importance scoring comprehensively considers the social influence of the event, the public attention, and the degree of association of the event in the knowledge graph; S55, screening the event set according to the importance score, selecting several events with the highest scores as core events for summary generation, and generating preliminary content of the news summary; S56. Based on the contextual information of the selected core events, natural language generation technology is used for further integration to generate an enhanced news summary containing key information.

6. The method for intelligent news broadcasting based on artificial intelligence according to claim 1, characterized in that: The S6 specifically includes: S61, assigning initial weights by analyzing the title, text and key sentence data in the enhanced news summary, and adjusting the weights according to the importance, timeliness and relevance of the news content; S62, processing the title data, and dynamically adjusting the weight of the title based on the keywords in the title and the relevance in the event knowledge graph; S63, further processing the text data, by analyzing the contextual relationship and paragraph importance of the text, and using a hierarchical attention mechanism to assign weights to different paragraphs or sentences in the text; S64, processing the key sentence data, and dynamically adjusting the weight of the key sentence according to the position of the sentence in the entire news, the context, and the importance of the explanation of the event background; S65. Introduce time-sensitivity optimization rules to dynamically adjust the news broadcast order of each part of the content according to the occurrence time, duration and immediate impact of the event on the user; S66. Apply the optimized weights and broadcast order to the final news broadcast script.

7. The method for intelligent news broadcasting based on artificial intelligence according to claim 1, characterized in that: The S7 specifically includes: S71. Performing logical consistency detection on the dynamically adjusted news broadcast content, using natural language processing technology to analyze the logical structure and semantic association of the news content, and identifying potential logical contradictions or inconsistencies; S72. Calculate the semantic similarity between paragraphs based on the semantic association of the event knowledge graph: ； in, Indicates the overall logical conflict degree, represents the semantic similarity between paragraph m and paragraph n, represents the association strength between event m and event n in the event knowledge graph, M represents the total number of paragraphs, and Q represents the total number of events; S73, using a credibility scoring mechanism to evaluate the credibility of the identified logical conflict paragraphs, and deciding whether to automatically resolve them based on the scoring results; S74. Combine the semantic association of the event knowledge graph to automatically select the most appropriate alternative content to resolve the conflicting paragraphs: ； in, Indicates alternative information, represents the set of all candidate nodes related to paragraph m, represents the weight of candidate node j, represents the semantic similarity between candidate node j and paragraph m, and R represents the number of candidate nodes; S75. Score the logical consistency of the digested news content, and generate and output logically consistent optimized broadcast content.

8. An artificial intelligence-based news intelligent broadcasting system, which executes the artificial intelligence-based news intelligent broadcasting method according to any one of claims 1 to 7, characterized in that: Includes the following modules: News data acquisition module, used to collect data and generate structured event data; The event knowledge graph generation module is used to generate an event knowledge graph based on structured event data, perform cluster analysis, and generate priority tags; A dynamic broadcast script generation module is used to generate dynamic broadcast scripts through a multimodal generative adversarial network; Emotion analysis module, used to analyze the user's voice signal and facial expression data to obtain the emotional state; Personalized speech synthesis module, used to generate personalized speech broadcast content according to user emotions; A news summary generation module, used to generate enhanced news summaries; The broadcast content sorting and optimization module is used to adjust the order of news broadcasts and assign weights; Logical consistency detection module, used to detect and resolve logical contradictions and generate optimized broadcast content; The news broadcast content adjustment module is used to adjust and update the news broadcast content in real time according to the user's real-time voice command data or the user's gesture signal data.

Images