Analysis of sentiments using unsupervised learning techniques

This paper proposes a new model called Combined Sentiment Topic (CST) model to detect sentiments and topics simultaneously from text based on Gibbs sampling algorithm, which performs better compared to existing semi- supervised approaches.

Abstract

Sentimental analysis, a sub discipline within data mining and computational linguistics, refers to the computational methodology for mining, understanding and assessing the opinions expressed in many opinion rich resources like blogs, discussion forums etc. The goal of sentiment analysis is to identify emotional states in online text. Most of the time classifiers trained in one domain do not perform well in another domain. Also the problem in existing approaches is not to detect sentiment and topics. Sentiments may vary with topics. This paper proposes a new model called Combined Sentiment Topic (CST) model to detect sentiments and topics simultaneously from text. This model is based on Gibbs sampling algorithm. Besides, unlike supervised approaches to opinion mining which often fail to produce good performance when shifting to other domains, the unsupervised nature of CST makes it highly portable to other domains. CST model performs better compared to existing semi- supervised approaches.