A Pair-Wise Method for Aspect-Based Sentiment Analysis

摘要

Aspect-based sentiment analysis aims at identifying the sentiment polarity of specific target in its context. Researches mainly focus on the ways for exploring the sentiment polarity based on explicit aspect of different products. The existing approaches have realized the sentiment classification with given targets and developed various methods with the goal of precisely polarity classification. However, there are no given explicit aspects in most practical scenarios. In this paper, we propose a pair-wise method which merges aspect-sentiment pair extraction and polarity classification in an unified framework. We convert the aspect-sentiment pairs detection process into a pairs binary classification problem correspondingly. Meanwhile, we construct a feature system applied to opinion mining. The experimental results on CCF BDCI2017 aspect-based sentiment analysis shared task dataset show that our proposed pair-wise method obtained good performance with a 0.718 F1 score which outperforms most proposed methods.