Research On Text Sentiment Analysis Based On Deep Learning

With the development of the internet,text data has become one of the main ways for people to interact online.These data contain not only information from various industries,but also user sentiment,which can reflect attitudes towards events of people.Therefore,text sentiment analysis has become a hot topic in natural language processing.However,the diversity of sentiment words and categories in sentences can hinder the performance of sentiment analysis,which is a challenge in the field.To address the problem of inaccurate sentiment judgment caused by the mutual influence of different sentiment words in sentences,a text sentiment binary classification algorithm based on the Sentiment Dictionary and Graph Attention Network is proposed in this paper.Textual semantic features are extracted through the RoBERTa model,and a sentiment graph network is constructed using whole sentences,clauses,and sentiment words in the sentiment dictionary as nodes.Weighted encoding of each node is achieved through Graph Attention Network,and the whole sentence nodes containing all information are concatenated with the semantic features of the input text to obtain classification results via a classifier.The simulation results show that the accuracy,precision,Recall,and F1 of the proposed algorithm are superior to other algorithms.To address the problem of ineffective utilization of label information caused by excessive differences between labels and text,a multi-label text sentiment classification algorithm based on Label Information Enhancement and Text Heterogeneous Graphs is proposed in this paper.The RoBERTa-WWM is used to encode the text and labels separately.The resulting representation is then constructed as a "word-label" heterogeneous graph and encoded through Graph Attention Network weighting to update the representation of label nodes.The final label features and text features are weighed via the attention mechanism to obtain classification results.The simulation results show that the multi label text sentiment classification algorithm proposed in this paper outperforms other algorithms in both single sentiment classification and overall multi label classification performance.The jaccard,micro F1,macro F1,Accuracy,and F1 indicators are higher than other algorithms,indicating better robustness.To achieve the application of sentiment classification in Chinese text comments on the Internet,this paper designs a sentiment analysis system.The system has user management functions and implements automatic crawling and sentiment analysis of online comment texts.