Research On Sentiment Analysis Of Financial Forum Text Based On Dependency Parsing

The rapid development of Internet technology provides an open platform for people.More and more people are inclined to discuss current affairs and share their ideas on the Internet.This has resulted in an increasing number of information resources on the Internet.At the same time,due to the wide coverage of the Internet and the high degree of user participation,a large number of information resources generated from interaction has become an important basis for decision-making of managers in various fields.The full exploitation of these resources will help decision-makers to grasp the direction of development.This is especially true for investment activities in financial market.The information asymmetry may bring great risk,and is not conducive to the stable and healthy development of financial market.In the financial forum,a large number of investors have gathered,and they have formed a large amount of unstructured text data.The emotions and attitudes contained in these data are great significance for both the regulators and investors.Therefore,it is necessary to design sentiment analysis algorithm for this text.Sentiment analysis is a hot and difficult problem in Natural Language Processing.In this study,first of all,based on the relevant research,this paper constructs the basic resources of emotion analysis,such as emotional lexicon,corpus,etc.We design the sentiment analysis algorithm mainly based on the analysis results of dependency.Taking into account that the subject,predicate and object are the main components in the Chinese grammar,this paper puts forward the concept of emotional backbone.On the basis of the artificial experience,the paper sums up some rules of affective computing,and puts forward the SPO sentiment analysis algorithm.On this basis,this paper analyzes the rules of emotion calculation between words,and introduces the interaction terms in the computational model,so as to be able to describe the interaction between words.On the estimation of the parameters of the model,a group of training data are manually labeled and the parameters are estimated by genetic algorithm.Based on this,this paper proposes a GA-SPO sentiment analysis algorithm.The experimental results show that SPO algorithm compared with the traditional machine learning algorithm is significantly improved in each evaluation index,especially in negative text.The role of syntactic information in the analysis of Financial Forum text is demonstrated.Meanwhile,compared with the SPO algorithm,GA-SPO algorithm has better performance in the same test set.