A Hybrid HDP-ME-LDA Model for Sentiment Analysis

Abstract

Abstract. Sentiment Analysis is an important research area in data mining. Recently, various topic models for aspect detection and sentiment analysis have been proposed, many of which are based on Latent Dirichlet Allocation(LDA) unsupervised machine learning approach. LDA requires the number of topics in advance which are often based on unreliable experience for different areas. On the other hand, it is important to identify aspect and opinion words of topics, especially aspect-specific opinion words and analyze sentiment polarity. But few research did them all. To solve these problems, this paper proposes a hybrid Hierarchical Dirichlet Process and Maximum Entropy-Latent Dirichlet Allocation(HDP-ME-LDA) model. It uses HDP to automatically determine the number of topics and utilizes maximum entropy classifier to separate aspect and opinion words, including aspect-specific opinion words. We evaluate our model on data sets of reviews of restaurant and electronic devices qualitatively and quantitatively. The result shows that we perform better than other topic models, like JST, ASUM, MaxEnt-LDA, HDP-LDA.

Comparative Study on Perceived Trust of Topic Modeling Based on Affective Level of Educational Text

“…Consequently, the major contribution of this study is to lay a foundation for a transition of prevailing technical viewpoints in the integration of LDA and SA to a user viewpoint. Indeed, there are various studies addressed the integration of LDA and SA through an improved semantic algorithm for LDA [38][39][40]. However, existing studies lack important discussion for how users perceive information delivered from LDA and SA.…”