Design And Implementation Of Social Network Hot Topic Sentiment Analysis System

With the continuous development of Internet technology,the rapid increase in the number of Internet users and the rapid spread of mobile devices,there are a large number of online users on social networks who express their opinions and express their feelings on hot topics.The emotional tendency of hot topics has high commercial value,political value and social value.In the social network,such as the content on Weibo,the language style is free and new words appear frequently,which is very different from the traditional hot topic discovery and sentiment analysis.Therefore,this paper studies the sentiment analysis tasks of social network hot topics from three aspects: new word discovery,hot topic discovery and sentiment analysis.The specific work is as follows:Firstly,this paper studies and optimizes the new word discovery method of statistic combined with filtering rules.The statistical method is used to measure the solidification degree and freedom degree of candidate new words,and various filtering rules are developed to improve the effect of new word discovery.Experiments have shown that statistics combined with filtering rules method have better results than statistical-based method and rule-based method.Secondly,this paper proposes a hot topic discovery method enhanced by semantic matching model.The method mainly includes short text clustering method enhanced by semantic matching model(STCSMM)and topic keyword extraction based on TF-IDF.STCSMM learns the text representation and vector distance calculation in clustering by training the semantic matching model.Experiments show that the hot topic discovery method enhanced by the semantic matching model can achieve more effective hot topic discovery.STCSMM has achieved the best results in the clustering effect evaluation indexes(Jaccard coefficient and FM index),and the topic keyword extraction method based on TF-IDF is more accurate and efficient.Thirdly,this paper designs a two-channel feature model based on BERT sentence vector for sentiment analysis of texts in social networks.The model combines BERT sentence vector,Bi LSTM and attention mechanism as a two-channel structure,which combines character level embedding and word level embedding,Transformer type features and RNN type features to realize the sentiment classification of text.The validity of the two-channel feature model based on BERT sentence vector is proved by experiments.This method has better performance than other comparison methods in the evaluation index F-measure.Finally,this paper designs and implements a social network hot topic sentiment analysis system based on the research above.By acquiring user information and content on the social network platform,the system performs new word discovery,hot topic discovery,sentiment analysis and the like on the data,and visually displays the result to the system user.The system acquires user information and content on the social network platform,processes and analyzes the data,such as new word discovery,hot topic discovery,sentiment analysis,and visually displays the results to system users.The system can discover new words appearing in social networks,focus on hot topics,and get the results of sentiment analysis to help commercial organizations and government departments make decisions.