Application of Aspect-based Sentiment Analysis using Self-Attention Mechanism to Japanese Sentences

Abstract

Sentiment analysis is a task to estimate emotions from information such as sentences. As SNSs such as LINE and Twitter which communicate mainly using sentences have developed, technology to estimate emotion from sentence information is increasingly important. In recent years, a method using aspect information has attracted attention as a sentiment analysis method using context information. The aspect-based sentiment analysis is realized in three stages. It categorizes sentences at the first stage, estimates word position which is a concrete description of aspect at the second stage, and outputs the polarity at the last stage. In this research, a neural network-based method using the self-attention mechanism for performing the aspect-based sentiment analysis is applied to Japanese sentences to evaluate its performance in comparison with English sentences.