Hotel reviews sentiment analysis based on word vector clustering

A text feature vector representation method based on Word2Vec and ISODATA clustering algorithm can solve the problem that word clustering result is sensitive to the initial number of clusters and, moreover, accuracy of text sentiment classification will be improved.

Abstract

In traditional text sentiment analysis methods, text feature vector has the problem of high dimensionality and high sparseness. In view of this situation, we can cluster the similar words together and use the generated clusters to fit into a new dimension so that the text feature vector dimension will be decreased. By using Word2Vec tool and K-means clustering algorithm, this task can be completed. However, clustering result of K-means clustering algorithm is sensitive to initial number of clusters. Therefore, we propose a text feature vector representation method based on Word2Vec and ISODATA clustering algorithm. This method can solve the problem that word clustering result is sensitive to the initial number of clusters. By using this method, text feature vector can be represented better, moreover, accuracy of text sentiment classification will be improved. The experimental results show that, under different initial number of clusters, the accuracy of our method is about 0.25% higher than that of text feature vector representation based on Word2Vec and K-means clustering method in hotel reviews sentiment analysis, and the AUC value is increased by about 0.31% on average.