Learning the Mapping Rules for Sentiment Analysis

Abstract

There is an increasing popularity of people posting their feelings on microblogging such as Twitter. Sentiment analysis on the tweets allows organizations to monitor public' feelings towards a product or brand. In this paper, we model sentiment analysis problem as a multi-classification approach that utilizes various feature types, including predicate-argument relation, hashtag, mention and emotion in the tweets. We describe a Content-Structure Correspondence (CSC) model that is able to represent diverse feature types in a tweet. We present a conceptual hierarchy to express the characteristics of a tweet. A multi-classification approach is used to map tweet content to the conceptual hierarchy. The mapping patterns are learned to identify the sentiment of a tweet.