User Clustering Based on the Topic-Sentiment Analysis

Nowadays, more and more people communicate and comment on what they care about on social platforms. In order to determine the distribution of topic-sentiment in microblog comment text more accurately and efficiently, this paper combined the improved TF-IDF parallel algorithm and Latent Semantic Model to construct the topic-sentiment model for user clustering. Firstly, we carried out data pre-processing and semantic role tagging. Then, we used the improved TF-IDF parallel algorithm to calculate the eigenvalues and sentimental polarity values of each sentimental word, and then the topic-sentiment words table was established. Finally, we established a Latent Semantic Model for the data of users' topic sentimental polarity value, extracting eigenvectors of users, and clustering users with the same eigenvectors by K-means algorithm. Compared with the traditional models, the result of the experiment showed that the proposed model has better accuracy, recall rate, F1 and time performance.