Topic Mining And Knowledge Graph Construction For Emergencies Based On Social Media

With the advent of the Web 2.0 era,social media is also more and more popular.Whenever an emergency occurs,many users choose to post their opinions on social media platforms such as Sina Weibo and Twitter,forward and disseminate information related to the event,fusing a lot of irrelevant information to form massive social media data.And how to extract a hotspot emergency and related topics from a large amount of data and present it to the users in an easy-to-understand manner is a direction worthy of research.However,the short text,irregularity,high redundancy and large size of social media data make traditional topic models and event detection methods inapplicable to social media data.Therefore,this paper proposes a new method for the event detection based on the co-occurrence network,which is more suitable for the topic mining of social media short text.Combined with this method to detect emergencies,the events and the information corresponding to the events are visualized in the form of knowledge graph to improve the understandability and traceability of the events.The main research work of this paper is as follows:(1)A method for mining topics of emergencies based on co-occurrence network is proposed.This method uses TFIDF to extract feature keywords from social media texts that can reflect the topics of the event.Based on the co-occurrence relationship of the keywords between texts,the keywords are used to construct a co-occurrence network.The Louvain algorithm performed on the co-occurrence network,the communities are detected and the topic communities are identified.Taking the 2012 Beijing storm Weibo data as an example,the keywords of the storm disaster event topics extraction,network construction,community detection,and topic identification were performed,and the LDA model was used to extract the storm disaster topics.The comparison of the subject results of the two methods proves the effectiveness and superiority of the proposed method on social media data.The distribution of disaster-related topics in space and time was also analysed to provide important support for disaster assessment and understanding and decision-making of emergency measures.(2)Build an event knowledge graph based on social media.Event is detected based on the topic mining method proposed in this paper,combined with named entity recognition technology to obtain event time,place,people,text and other entities.And the relations between the entities is constructed in the form of triples.The relationship is stored in a graph,so that events are regularly visualized in the form of knowledge graph,and the intelligibility of the event is improved.Based on the Twitter data of Hong Kong events in 2019,one day is used as a unit of time to extract events and event-related entities throughout the whole time interval.A knowledge graph is built and news are referred to prove the authenticity of the extracted events.Based on the knowledge graph,single-day hot event analysis and quick retrieval of event knowledge can help understand the relevant information of the event in detail,so as to strengthen the understanding of Hong Kong event throughout the time period.