| With the rapid development of the Internet,network media has undergone earth-shaking changes.The way people get news has changed from the original paper media,such as newspapers,books,periodicals,etc.,to the Internet media.As the scale of online media continues to increase,the amount of news on the Internet also increases exponentially.Faced with the overwhelming variety of news reports,it is difficult for users to grasp the context of events,and discover the topics they want to see from the massive news reports,as well as grasp the current hot topics and events.In order to facilitate users to quickly grasp the current network media news in a short period of time,and to clarify the relationship between various events,it is necessary to conduct in-depth analysis and mining of news data,and extract hidden information from various news data.This topic conducts in-depth analysis and mining of news data in online media to facilitate users’ rapid extraction and enrichment of network information.Based on this goal,the following researches were carried out:1)Propose an evolutionary graph construction algorithm that incorporates syntactic dependencies.In the event relationship extraction module,this topic proposes a novel structure based on the syntactic dependency analysis tree:the event dependency graph,and designs a BEGC model based on the event dependency graph that jointly extracts the event causality and timing relationship.This method emphasizes the dependence of each meta-event in the sentence,and the model performs experimental evaluation on the news corpus and performs well.2)Propose a subject event detection model based on the evolutionary graph.Combining the evolutionary graph and GCN,the evolutionary Graph-GCN model is designed,which uses the structure of the evolutionary graph and the relationship between text words to extract text features and learn the news text representation.This method combines the structural features of the graph and the relationship features between text and words(text and word,word and word)to learn text representation vectors and perform event discovery tasks.A large number of experiments on the data set show that our model is superior to mainstream models in F1 and NMI indicators.3)Propose an aspect-based hot event detection algorithm.Considering the characteristics of news,normalizing the impact indicators of news events,selecting hot events in various fields,and providing different screening results for users with different interests.A large number of experiments on the data set show that our model can make the experimental results evenly distributed in various fields under the premise of ensuring the accuracy and recall rate,and meet the needs of users with different points of interest,and has excellent performance in the number standard deviation.4)Designed and developed an integrated news event analysis system based on the affair map.It integrates multiple algorithm components and provides a friendly interactive interface.It can support users to search for related news through keywords and view meta-events of related news,view the affair graph generated by related news data,and view thematic events of related news and internal events in a certain field.. |
