Research On Construction Technolongy And Application Of Knowledge Graph For Internet Public Opinion Event

With the rapid development of social networks,more and more netizens pay attention to social hot public opinion events.The active participation of netizens in the discussion of social hot public opinion events leads the rapid spread of online public opinion.The vast amount of events contains a wealth of information,thus effective management and analysis of those events will help to grasp the network public opinion situation.The existing research is not detailed enough and inadequate in the depth of mining information.There are two main problems:(1)Most of the organization and management of social hot public opinion events are using relational databases,which makes query retrieval difficult and fails to efficiently visualization and inference.(2)The analysis of propagation and diffusion based on social hot public opinion events is mostly based on simulation methods,which are complicated in theory and difficult to operate.The core of solving these problems lies in improving the storage and representation of network public opinion events,and then applying the propagation and diffusion of network public opinion events on the basis of the theory above.Knowledge graph,a new type of data representation and storage tool based on graph model,can effectively solve the problems.This thesis studies construction technolongy and application of knowledge graph for Internet public opinion event.The main research results are as follows:(1)Entity disambiguation method is explored.Aiming at overcoming the problems of weak semantic expression capabilities of traditional entity disambiguation methods and the lack of consideration of local characteristics of entities,an entity disambiguation method based on context word vectors and topic models is proposed.Firstly,the context word vector is trained by adding the context direction vector to the traditional word vector model.Secondly,the context similarity,category topic similarity,and topic similarity are calculated respectively.Finally,the three similarities are fused,and the entity with the highest similarity is selected as the disambiguated entity.Experimental results show that this method achieves better performance than current mainstream entity disambiguation methods.(2)Knowledge representation and reasoning method is studied.Aiming at resolving the ignorance of the rich information of neighbor entities,a knowledge representation and reasoning method based on attenuated graph attention networks is put forward.Firstly,the attenuated attention mechanism is introduced so that the weights of different relations are assigned to the different relation paths,which are combined with the graph attention network to design a graph attenuated attention networks.Secondly,the relation and entity representations are embedded,and the hidden features of the triples are mined on the basis of the network.Finally,based on the embedding of entities and relations,link prediction,relation prediction,and triple classification tasks are performed to complete knowledge reasoning.Experimental results show that this method outperforms mainstream knowledge representation methods based on neural networks.(3)The propagation and diffusion analysis method of Internet public opinion events is researched.Aiming at solving the problems of complex theoretical analysis and difficult operation of methods based on simulation and deduction,a knowledge graph based network public opinion event propagation and diffusion analysis method is presented.Firstly,a knowledge graph of network public opinion events is constructed.Secondly,user influence based on the knowledge graph is calculated,and opinion leaders are mined.Then,the Internet public opinion event propagation model is constructed based on the two-level communication theory to characterize online public opinion.Finally,diffusion forecast technology based on knowledge reasoning is designed,which are used to complete diffusion forecast through three steps: link prediction,relation prediction,and triple classification.The experimental results show that this method can effectively calculate the user influence,and then can effectively mine opinion leaders,and perform better propagation analysis and diffusion forecast on Internet public opinion events.