Sentiment Classification Using Information Extraction Technique

Abstract

This paper explores the sentiment classification with Information Extraction (IE) approach. The IE approach here is required to detect the sentiment expressions on specific subject (person, product, company and so on) and then to evaluate the sentiment strength and/or the validation of them. Our method can be illustrated logically as: (1) From a given text, extract the sentiment expressions on the specific subjects and attach certain sentiment tag and weight to each of them; (2) Calculate the sentiment indicator for each sentiment genre by accumulating the weights of all the expression with the corresponding tag; (3) Given the indicators on different sentiment genres, use a classifier to predict the sentiment label of the given text. To extract expression robustly when encounter some complex linguistic phenomena (such as ellipsis, anaphora), a new parsing idea named super parsing is proposed. It enables some non-adjacent linguistic constituents to be merged to deduce a new one. As an incremental implementation of super parsing, a system named Approximate Text Analysis (ATA) is described in this paper. As for the classification task, two different classifiers are used: simple linear classifier (called SLC here) and SVM. The experiments show the reasonable performance of our approach.