Social Media Sentiment Analysis Based On Knowledge Graph

With the rapid development of online social platforms,social media has also emerged recently.Social media publishes a massive amount of infomation every day,and the topics involve various fields.The emotional information implied in the content of social media has a great value for public opinion monitoring and polling.However,the current sentiment analysis technology is mainly for general text,but still has the following problems when applied in social media scenes: First,the diversified and specialized characteristics of social media content make it difficult for the sentiment analysis model to fully understand social media information without the support of prior knowledge.Second,the traditional sentiment analysis algorithm has a relatively coarse granularity,making it difficult to deeply explore the causes of netizens' emotions.Third,social media generates large amounts of fragmented information everyday,and traditional sentiment analysis systems lack effective organization and management of social media content.It is difficult for users of sentiment analysis systems to quickly obtain key information such as social media content entities,emotional tendencies,and emotional causesFor the problem that traditional sentiment analysis methods are difficult to fully understand social media information due to the lack of prior knowledge and the methods are coarse-grained,this thesis proposes a social media sentiment analysis method based on Uniform Content Label(UCL)and knowledge graph.This thesis uses Wikipedia corpus to design a Wikipedia Knowledge Graph(WIKIKG)and proposes a Wikipedia Knowledge Extraction Algorithm(WIKI-KEA).After the construction of WIKIKG is completed,a sentiment analysis algorithm named KGBSA which ultilizes knowledge graph for social media is proposed.This algorithm can provide users with fine-grained social media sentiment analysis results.The main work of this thesis is as follows:1)Aiming at the problem that traditional sentiment analysis methods are difficult to fully understand social media information due to the lack of prior knowledge,this thesis proposes a knowledge graph construction method based on Wikipedia.Firstly,we use crawler technology to crawl Wikipedia web content,extract structured data and unstructured text,and use UCL to index Wikipedia content.Then the UCL structured data is extracted to construct the WIKIKG basic library,and we design a knowledge extraction algorithm WIKI-KEA based on deep learning technology.This algorithm can extract entity relationship triplets from UCL unstructured text.The triplets are stored in WIKIKG after disambiguation.Finally,in the view of the characteristics of fast update of social media content,a real-time update method of knowledge graph based on social media is proposed to ensure that WIKIKG provides real-time and accurate prior knowledge to sentiment analysis system.2)Aiming at the problem of lack of effective organization and management of social media content and the problem of coarse-grained existing sentiment analysis methods,this thesis proposes a UCL-based content indexing method of social media content and a social media sentiment analysis algorithm called KGBSA.Firstly,the UCL-based social media content indexing method is designed to index key information in social media content.Then,we use prior knowledge in the knowledge graph,combined with social media features to build sentiment dictionaries.Finally,KGBSA algorithm is proposed,which is divided into two parts: KGBSA-ECD emotion cause detection algorithm and KGBSA-SC sentiment classification algorithm.The KGBSA-ECD algorithm uses graph attention networks to fuse text-related knowledge,and combines semantic information and location information features to predict emotional causes.The KGBSA-SC algorithm uses sentiment dictionaries and syntactic dependency trees to calculate sentiment tendencies in a sentiment sentence,and then complete the sentiment sentence classification task.3)The comparative experiments of the WIKI-KEA algorithm,entity disambiguation algorithm and KGBSA algorithm are designed respectively.The experimental results show that the WIKI-KEA algorithm and the entity disambiguation algorithm have better performance on public datasets,and can support the construction of WIKIKG knowledge graph well.The KGBSA algorithm has higher accuracy than traditional sentiment analysis algorithms.Based on the above experiments,the overall framework of the social media sentiment analysis system based on knowledge graph is designed and implemented in the dual-structural network prototype system.