Collective Opinion Target Extraction in Chinese Microblogs

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1840–1850, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Collective Opinion Target Extraction in Chinese Microblogs Xinjie Zhou, Xiaojun Wan* and Jianguo Xiao Institute of Computer Science and Technology The MOE Key Laboratory of Computational Linguistics Peking University No. 5, Yiheyuan Road, Beijing, China {zhouxinjie,wanxiaojun,xiaojianguo}@pku.edu.cn Abstract Microblog messages pose severe challenges for current sentiment analysis techniques due to some inherent characteristics such as the length limit and informal writing style. In this paper, we study the problem of extracting opinion targets of Chinese microblog messag- es. Such fine-grained word-level task has not been well investigated in microblogs yet. We propose an unsupervised label propagation al- gorithm to address the problem. The opinion targets of all messages in a topic are collec- tively extracted based on the assumption that similar messages may focus on similar opinion targets. Topics in microblogs are identified by hashtags or using clustering algorithms. Ex- perimental results on Chinese microblogs show the effectiveness of our framework and algorithms. 1 Introduction Microblogging services such as Twitter 1 , Sina Weibo2 and Tencent Weibo3 have swept across the globe in recent years. Users of microblogs range from celebrities to ordinary people, who usually express their emotions or attitudes towards a broad range of topics. It is reported that there are more than 340 million tweets per day on Twitter and more than 200 million on Sina Weibo. A tweet means a post on Twitter. Since we mainly focus on Chinese microblogs instead of Twitter in this paper, we will refer to a post as a message. Each message is limited to 140 Chinese characters and usually contains several sentences. * Xiaojun Wan is the corresponding author. 1 https://twitter.com 2 http://weibo.com/ 3 http://t.qq.com/ Currently, researches on microblog sentiment analysis have been conducted on polarity classifi- cation (Barbosa and Feng, 2010; Jiang el al., 2011; Speriosu et al., 2011) and have been proved to be useful in many applications, such as opinion poll- ing (Tang et al., 2012), election prediction (Tumasjan et al., 2010) and even stock market prediction (Bollen et al., 2011). However, classify- ing microblog texts at the sentence level is often insufficient for applications because it does not identify the opinion targets. In this paper, we will study the task of opinion target extraction for Chi- nese microblog messages. Opinion target extraction aims to find the object to which the opinion is expressed. For example, in the sentence “The sound quality is good!”, “sound quality” is the opinion target. This task is mostly studied in customer review texts in which opinion targets are often referred as features or aspects (Liu, 2012). Most of the opinion target extraction approaches rely on dependency parsing (Zhuang et al., 2006; Jakob and Gurevych, 2010; Qiu et al., 2011) and are regarded as a domain-dependent task (Li et al., 2012a). However, such approaches are not suitable for microblogs because the natural language processing tools perform poorly on …

Content truncated to 500 words