Research On The Stance Detection And Its Evolutionary Analysis Method Of Comments On Hot Topics

With the emergence of domestic and foreign news events and hot topics,more and more Internet users choose to express their opinions and comments on the topics.In the face of massive amounts of data,how to accurately and quickly discover hot topics plays an important role in many areas,such as national security,public opinion surveys,and business decision-making etc.At the same time,in the face of these hot topics,users often directly or indirectly express their attitudes toward the corresponding topic,such as "FAVOR" or "AGAINST",Analysis of their comments can reflect the user's position on the topic.Mining opinions and stance tendencies of comment texts is an important area of fine-grained sentiment analysis.This paper studies the key technologies of hot topic discovery,opinion extraction,stance detection and its evolution analysis.The specific research contents are as follows:1)A short text hot topic discovery method based on BTM and K-means is proposed.This method is based on the Biterm Topic Model(BTM),and introduces the topic word vector after obtaining the corresponding topic word,and improves the performance of the short text topic model by fusing the relationship between the word and the context.Finally,on the basis of traditional K-means,different feature weights are assigned to each attribute feature,which reduces the influence of noise dimensions on clustering,and then clusters the results to realize short text hot topic discovery.2)A method of extracting opinion based on POS-Bi-LSTM-CRF is proposed.This method first uses the Language Technology Platform(LTP)to tag the comment text.Then use the bidirectional long-short term memory network(Bi-LSTM)to fuse the char vector with its part-of-speech(POS)features to obtain the initial text representation.Finally,the text representation is processed through a neural network layer such as Bi-LSTM and Conditional Random Field(CRF)to achieve evaluation objects and opinion extraction.3)A stance detection and Its evolutionary analysis method based on IAN-Bi-LSTM is proposed.This method uses deep learning methods to achieve stance detection classification.The topic word vector and the comment text word vector are input into the model at the same time,and then the stance detection of the comment text is realized after the neural network layer such as Bi-LSTM and interactive attention mechanism.Finally,the comment text is divided into time slices,and the stance evolution of the comment texton different time slices is analyzed.The experimental results prove that the three models proposed in this paper have achieved good performance.For short text topic review data,BTM combined with word vector features and improved K-means algorithm can well complete short text hot topic discovery to a certain extent;the F value of the review opinion extraction method based on the deep learning model reaches about 88%;The deep learning method strengthens the relationship between the review text and the topic,the accuracy of the position orientation classification is improved,and then analyzes the evolution of the stance of the user reviews on the time series.Finally,it explains the problems in the research work of this article and the next research work.