Design And Implementation Of News Public Opinion System Based On Business Domain Knowledge Graph

With the development of science and technology,the main battlefield of public opinion analysis has shifted from traditional media to online media represented by forums,online news,and Weibo.These online media mainly exist in the form of texts,which have the characteristics of strong information bursting,fast spreading speed,strong one-sidedness,and bias in reporting.They have a significant influence in the business field,especially the news published by authoritative media,have an impact on business or personal decisions.Based on the needs of the business field,this paper proposes a news public opinion system based on the knowledge graph of the business domain.The system components include a crawler system,a knowledge graph,and a data mining module.The data mining module is composed of an entity linking and a sentiment analysis module.The main contributions and innovations of this paper are as follows:(1)In order to ensure the accuracy of the commercial public opinion analysis from the text recognition entity,and to solve the traditional methods based on support vector machine or PageRank need to manually extract features,long construction period,high labor costs,this paper proposes an Entity Linking method based on knowledge graph.First,the method establishes a knowledge graph of the business domain,which contains the names,attributes and relationships between the entities.Then,based on the information of the entities,the method uses the neural network model to implement the link of the text to the knowledge graph.(2)In order to ensure that the public opinion system correctly recognizes the influence of long texts such as news on the entity,and solves the traditional problem of low accuracy of emotion analysis for specific entities based on support vector machine SVM and logistic regression,this paper proposes a sentiment analysis model for specific goals of long texts.The method first finds the text content of the sentence-related text related to a specific entity,and then obtains the vector representation of the specific entity in each sentence,and uses the attention mechanism to combine to obtain the overall representation of the specific vector,and finally uses the overall representation to perform the emotion analysis.(3)According to the characteristics of fast updating of information in the business field,we use git tool to manage XML files to store and maintain the knowledge graph of more than 10,000 entities.(4)Deploy and launch the system proposed in this paper.The system crawls 3,000 daily news on average and has enterprise users such as Evergrande,Country Garden and Interface News.