Research On Sentiment Analysis Method Based On Multi-algorithm Collaboration

With the development of social media and the growth of Internet users,network information has shown explosive development,especially the mainstream social platforms represented by Weibo,which have become an important channel for the general public to express their emotions and opinions.People share their daily life and express their views on various events on these platforms.In this context,paying attention to Weibo users’ emotional tendencies towards events will help the platform understand users’ voices,and can also provide reference and direction for decision makers in handling public opinion.However,most of the current microblog sentiment analysis research is still based on text,with insufficient consideration of elements such as expressions and pictures,and no mining of the reasons for user emotions.In response to the above problems,this article proposes a multi-feature topic sentiment analysis model with higher sentiment classification accuracy.Firstly,relevant microblog data is crawled and preprocessed according to a given topic,and then input into the improved Senti-BERT model in this paper.Perform text feature representation,and then extract more complete context features through Bi LSTM-Attention,and combine the influence of emoji emoticons in emotional expression,and finally complete the emotional classification task.On this basis,this article proposes an emotional reason pair extraction model based on deep learning to deeply mine the reasons for user emotions.By introducing the Bi LSTM model with a multi-layer attention mechanism,the emotion and reason candidate sets are extracted in combination with context information,and the result of emotion prediction is used as part of the input for reason extraction to increase the interaction between emotion and reason.At the same time,considering the influence of relative position in the emotional reason combination,the relative distance is obtained through the relative position feature extraction network,and finally the matching of emotion and reason is completed.Experiments show that the experimental results of the two models proposed in this article have been improved to a certain extent in their respective tasks.The main research and innovation points of this article are as follows:(1)A calculation method for emoji expression features is proposed.First,vectorize the emoji expression,construct the emoji expression network graph,pay attention to the emotional connection between the emoji symbols,and then give different weights through the attention mechanism,integrate the text information features,and obtain the result through mapping.(2)A multi-model fusion Senti-BERT model is proposed.Integrating the emotional dictionary,first extract the emotional words and opinion words to the beginning of the sentence,and combine the [CLS] bits,hidden vectors corresponding to the emotional words and opinion words to calculate when obtaining the emotional label.And integrate with BERT-WWM and Ro BERTa,and use their advantages in Chinese text to improve the performance of the BERT model.(3)In the emotional reason pair extraction task,a multi-layer attention mechanism and relative position features are introduced,and the relative distance is obtained by using position coding.Considering the influence of relative distance in the combination of emotional reasons,the matching accuracy is improved.