Research On The Sentiment Classification And Influence Of User In Financial Social Network

With the development of the Internet,stock market investors are increasingly inclined to obtain financial information and exchange investment views through online platforms.The network platform contains rich financial data and investor comment information,which contains users' emotional tendency towards the market and has become an important factor influencing investors psychology and behavior.The classification of comments made by users in financial social networks can reflect the emotional tendency of users.And the comments made by users with greater influence in the stock market not only affect the judgment of ordinary shareholders,but also affect the trend of individual stocks and even the broader market to a certain extent.In order to quickly and accurately grasp the emotional fluctuations of the financial and securities markets,it is necessary to classify stock reviews to obtain investor users' long and short view of the market,at the same time,by comparing the opinion results with the trend of the overall market,the prediction accuracy characteristics of the stock rise and fall were constructed,and combined with its own characteristics,the user influence is calculated and ranked.With the continuous development of deep learning,more and more people are beginning to apply deep learning to the study of sentiment classification.However,the texts in financial social networks have problems such as colloquialism and less content,which leads to the problem of information redundancy and sparse features.Many existing emotion classification methods cannot fully consider the problems of information redundancy and feature sparseness.As for the research on user influence,traditional social network user influence calculation is generally based on user attributes or behaviors,and it lacks the features constructed by combining specific domain knowledge and a reasonable method to distinguish the importance of the constructed features.In view of the problems in the research of emotional classification and user influence of the above financial social networks,this paper mainly does the following work:(1)This paper proposes a BA-CNN-LSTM(BERT & Attention based CNN-LSTM)sentiment classification model.The model comprehensively considers the problems ofinformation redundancy and feature sparseness.Based on the Convolutional Neural Networks and Long Short-Term Memory fusion model,BERT(Bidirectional Encoder Representations from Transformers)model is used to replace the previous word vector conversion models.It comprehensively considers information such as word vectors,text vectors,and position vectors,and uses a Multi-head Self-Attention mechanism for pre-training tasks inside the model to obtain a semantic representation of the text containing rich semantic information.In addition,Attention is introduced into the hidden layer of Long Short-Term Memory to assign different weights to features,so that the model can focus on important words,so as to solve the problem of information redundancy in emotion classification research.Finally,a number of comparative experiments were carried out in this paper.The experimental results show that the sentiment classification model proposed in this paper is better than other comparison models.(2)This paper proposes a new method to calculate user influence.The method first determines the weight of the user feature by subjective and objective comprehensive weighting method,and then uses the weighted summation method to calculate the user influence.In addition,combined with specific domain knowledge,this paper constructs the characteristics of user influence calculation,such as user interaction frequency,user activity,user attention and user prediction accuracy.Through the comparison of experiments,it is verified that the weight determined by the subjective and objective comprehensive weighting method used in this paper is more reasonable than the weight determined by the subjective weighting method or the objective weighting method alone.(3)In this paper,the text sentiment classification results are compared with the market trend to construct the user prediction accuracy characteristics,and this feature is added to the user characteristics,and the influence calculation is combined with other characteristics of the user.