A Joint Model of Term Extraction and Polarity Classification for Aspect-based Sentiment Analysis

Abstract

Aspect-based sentiment analysis (ABSA) is a significant task in opinion mining, which aims to extract explicit aspects of an entity along with the sentiment expressed towards these aspects. To achieve this goal, two subtasks are performed: aspect term extraction (ATE) and aspect polarity classification (APC). However, recent work has solved these two subtasks separately or has only focused on either subtask. In addition, the sequential model of two subtasks may cause chain errors from ATE to APC and designing and running two models consumes too many resources. In this paper, we propose a joint model for ABSA that can deal with two subtasks, ATE and APC, simultaneously. The experimental results on two datasets from SemEval 2014 show that our model, which is named MATEPC (Model of Aspect Term Extraction and Polarity Classification), outperforms several baseline models in the ATE task and gives a promising result in the APC task by dealing with ATE and APC at the same time.