Constructing domain-dependent sentiment lexicons automatically for sentiment analysis

The sentiment lexicon is an important source for sentiment analysis, which has received lots of attention in recent years. But the word's sentiment inclination is often determined without taking into account the domain knowledge in most general sentiment lexicons. However, sentiments of some word are domain-dependent. Thus, the general sentiment lexicons, such as SentiWordNet, work with low performances in sentiment analysis applications. In this study, the problem of constructing a domain-dependent sentiment lexicon with supervised learning method was explored. The sentiment inclination of a word was identified by quantifying its relationship with the polarities of labels. Intensive experiments were carried out on a real dataset to show that the effectiveness of proposed approach, which was capable of detecting the word sentiment depending on a special domain correctly. Multiple sentiment classification tasks were performed for demonstrating the performances of the constructed lexicons, by which the classification accuracies were statistically improved significantly. © 2013 Asian Network for Scientific Information.