Evolution And Prediction Of Enterprise Online Public Opinion Based On Event Graph And Event Evolutionary Graph

As the main body of the market economy,enterprises play an important role in the market economy."Behaviors" of enterprises in the marketplace are complex,and the consequences thus produced can be tremendously different.Even one minor action may trigger public opinion and affect the whole business.Currently,the application of and research on enterprises' online public opinion data are mainly in two aspects.For one thing,it is used as a dimension of enterprise credit evaluation.Especially for many non-listed enterprises,it is difficult to obtain their real financial data and in most cases,public news is the only information available.For another,the research mainly focuses on the dissemination and emotion identification of public opinion.In-depth analysis and digging of public opinion contents are insufficient.Hence,there are two main problems in the current research on enterprise online public opinion.For one thing,most research concentrates on emotion analysis of commentary texts and public opinion dissemination mechanism,including the emotional tendency of texts as well as the origin,speed and scope of public opinion dissemination,while lacking indepth analysis of the current situation and development trend of public opinion,such as the semantic analysis and association analysis.For another,enterprises' online public opinion data is vastly large in quantity and mainly semi-structured,whose sorting,text analysis and semantic analysis require a lot of manpower,material resources and the support of text mining,machine learning and other technologies.Therefore,how to tap the current situation of enterprise operation and predict the development trend of enterprises by using the online public opinion news has become a hot issue of current research.Main contents of the current study are as follows.The first is to construct a research framework for enterprise online public opinion.Taking Le TV's online public opinion as an example,the significance of online public opinion is clarified for the first place,namely it reflects the status quo of business operation.The paper points out that enterprise online public opinion can evolve with time,form a structured network,and the public opinion texts are unstructured.It then goes on to analyze the influence of enterprise online public opinion,based on which two research objectives are determined,that is,present situation analysis and public opinion prediction.Centering on the objectives,the research contents and realization paths are identified and a framework for research on Event Graph(EG)and Event Evolutionary Graph(EEG)of enterprise online public opinion is constructed.The second is to research on EG construction of enterprise online public opinion based on semantic analysis.Firstly,named entities and dependency relations in the statements of public opinion texts are analyzed based on which public opinion events are extracted in the form of the subject-verb-object structure.Next,causality relations in texts are extracted by analyzing the forms and rules of causality in statements.Lastly,the extracted events and relations are matched to form the EG of enterprise online public opinion.The third is to research on the co-occurrence network of enterprise online public opinion keywords.Firstly,the keywords of enterprise online public opinion are extracted,the co-occurrence frequency of keywords is calculated,and the co-occurrence matrix is thereupon generated.Then,the social network analysis of the co-occurrence matrix is conducted through UCNET and NETDRAW,and the measurement indexes adopted are network centrality and K-core network of keywords,so as to analyze the influence of different subjects and events in enterprise online public opinion and recognize the core events in enterprise online public opinion.The fourth is to research on EEG construction of enterprise online public opinion based on machine learning.Firstly,subject words are transformed into vector patterns through the Word2 vec algorithm,and similar words are extracted to form a thematic lexicon of online public opinion through a million-level pre-trained news model,to provide pre-training data for subsequent automatic generalization of events.Next,a Naive Bayes classifier for event automatic generalization is constructed to realize the primary classification of extracted events.Again,the similarity of short texts is calculated through vectorization in order to realize the secondary classification,i.e.,event generalization.Finally,the transition probability among events in causality relations is calculated to generate the EEG of enterprise online public opinion which is then used to analyze specific cases.The innovation of this paper lies in the following aspects.First,an analysis model of enterprise online public opinion evolution is proposed based on EG.Firstly,EG is defined as the sorting and visual representation of enterprise crisis events and event relations behind online public opinion.Next,the semantic analysis of public opinion texts is conducted through natural language processing technology to extract public opinion events and causality relations,based on which the sequence and causality of crisis events are constructed and displayed in a visualized form.Second,a method to analyze the influence of enterprise online public opinion events is designed based on word co-occurrence network.Firstly,keywords of enterprise online public opinion events are extracted.Secondly,two word co-occurrence networks are constructed,one based on the co-occurrence of causality among keywords and the other on the co-occurrence of public opinion news.Thirdly,the influence of particular events in the evolution of enterprise public opinion is deduced by analyzing the influence of basic attributes of the network,word centrality and K-core on keywords.Finally,through comparison of the two co-occurrence networks,it is found that although the angle and focus of the two networks are different,they are highly related and can be complemented to each other in the analysis of the influence of public opinion events.Third,a predictive analysis model of enterprise online public opinion is constructed based on EEG.Firstly,the classification criteria for enterprise crisis events are determined through literature review.Secondly,the generalization of extracted events is realized based on machine learning algorithm and word vector similarity measure.Thirdly,the transition probability among events is calculated to form the enterprise online opinion EEG.Finally,the public opinion prediction of extracted cases is carried out based on EEG.There are 47 figures,54 tables and 161 references in this paper.