Opinion Words Extraction and Sentiment Classification with Character Based Embedding

Abstract

In recent years, sentiment analysis from customer comments has received widespread attention in deep learning and recognition computing area. In the field of fine-grained sentiment analysis, aspect level sentiment classification aims to detect the sentiment polarity towards a particular aspect in a sentence. Most of previous research in this task focus on sentiment polarity, and ignore the importance of opinion words. As the specific embodiment of sentiment, opinion words provide diversified representation of the aspect and contribute to sentiment polarity analysis. In this work, character level word embedding is applied to our model for enhanced semantic expression, and additional position attention based on opinion words is used in sentiment classification. Our work shows considerable improvement in opinion words extraction and acquires comparable results in sentiment polarity classification on SemEval 2014 datasets.