评价对象抽取研究综述

Survey on Opinion Target Extraction

近年来,细粒度情感分析因其在商业决策、舆情分析等领域的重要作用而受到学术界和工业界的广泛关注.评价对象抽取作为情感分析的基本任务之一,是进行细粒度情感分析的关键问题.本文针对评价对象抽取问题的起源、当前主流研究方法和趋势进行了梳理,首先详细阐述评价对象抽取问题的基本概念并对其进行形式化表示,然后结合近年来的研究对评价对象抽取方法进行归纳和总结,并重点分析基于频率、基于模板规则、基于图论、基于条件随机场和基于深度学习的评价对象抽取方法,随后回顾评价对象抽取的评测情况和可用的语料资源,最后分析评价对象抽取的若干难点问题,同时对评价对象抽取研究进展和发展趋势进行总结和展望.

In recent years, fine-grained sentiment analysis has received extensive attention from academia and industry for its important role in business decision-making and public opinion analysis. Opinion target extraction, as one of the basic tasks of sentiment analysis, has been the crux of fine-grained sentiment analysis for years. In this paper, the origins,state of the art, and research directions of opinion target extraction are discussed. We first elaborate the basic concepts of opinion target extraction and formalize the problem, and then based on recent literature, we conclude and summarize the approaches and techniques, which we divide into five categories： frequency-based method, pattern-based method,graph-based method, CRF-based method and deep learning based method. We also review several evaluation contests and collect the available corpus resources on opinion target extraction. At the end, the challenges arisen in opinion target extraction are analyzed as well as the probable future is given.