Sentiment analysis of movie reviews on discussion boards using a linguistic approach

A linguistic approach for sentiment analysis of message posts on discussion boards that generates a dependency tree, and splits the sentence into clauses, and determines the contextual sentiment score for each clause utilizing grammatical dependencies of words and the prior sentiment scores of the words derived from SentiWordNet and domain specific lexicons.

Abstract

We propose a linguistic approach for sentiment analysis of message posts on discussion boards. A sentence often contains independent clauses which can represent different opinions on the multiple aspects of a target object. Therefore, the proposed system provides clause-level sentiment analysis of opinionated texts. For each sentence in a message post, it generates a dependency tree, and splits the sentence into clauses. Then it determines the contextual sentiment score for each clause utilizing grammatical dependencies of words and the prior sentiment scores of the words derived from SentiWordNet and domain specific lexicons. Negation is also delicately handled in this study, for instance, the term "not superb" is assigned a lower negative sentiment score than the term "not good". We have experimented with a dataset of movie review sentences, and the experimental results show the effectiveness of the proposed approach.