A Generative Model for Identifying Target Companies of Microblogs

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 688–697, Dublin, Ireland, August 23-29 2014. A Generative Model for Identifying Target Companies of Microblogs Yeyun Gong, Yaqian Zhou, Ya Guo, Qi Zhang, Xuanjing Huang Shanghai Key Laboratory of Intelligent Information Processing School of Computer Science, Fudan University 825 Zhangheng Road, Shanghai, P.R.China {12110240006, zhouyaqian, 13210240002, qz, xjhuang}@fudan.edu.cn Abstract Microblogging services have attracted hundreds of millions of users to publish their status, ideas and thoughts, everyday. These microblog posts have also become one of the most attractive and valuable resources for applications in different areas. The task of identifying the main targets of microblogs is an important and essential step for these applications. In this paper, to achieve this task, we propose a novel method which converts the target company identification problem to the translation process from content to targets. We introduce a topic-specific generative method to model the translation process. Topic specific trigger words are used to bridge the vocabulary gap between the words in microblogs and targets. We examine the effectiveness of our approach via datasets gathered from real world microblogs. Experimental results demonstrate a 20.2% improvement in terms of F1-score over the state-of-the-art discriminative method. 1 Introduction With the rapid growth of social media, about 72% of adult internet users are also members of a social networking site1. Over the past few years, microblogging has become one of the most popular services. Meanwhile, microblogs have also been widely used as sources for analyzing public opinions (Bermingham and Smeaton, 2010; Jiang et al., 2011), prediction (Asur and Huberman, 2010; Bollen et al., 2011), reputation management (Pang and Lee, 2008; Otsuka et al., 2012), and many other applications (Bian et al., 2008; Sakaki et al., 2010; Becker et al., 2010; Guy et al., 2010; Lee and Croft, 2013; Guy et al., 2013). For most of these applications, identifying the microblogs that are relevant to the targets of interest is one of the basic steps (Lin and He, 2009; Amig´o et al., 2010; Qiu et al., 2011; Liu et al., 2013). Let us firstly consider the following example: Example 1: 11” MacBook Air can run for up to five hours on a single charge. “MacBook Air” can be considered to be the target being discussed on the microblog, and we can also infer from the microblog that it is related to Apple Inc. The ability to discriminate which company is being referred to in a microblog is required by many applications. Previous studies on fine-grained sentiment analysis and aspect-based opinion mining proposed supervised (Popescu and Etzioni, 2005; Liu et al., 2012a; Liu et al., 2013) and unsupervised methods (Hu and Liu, 2004; Wu et al., 2009; Zhang et al., 2010) to extract targets of opinion expressions. Based on the associations between opinion targets and opinion words, some methods were also introduced to simultaneously solve the opinion expression and target extraction problems (Qiu et al., 2011; Liu et al., 2012a). However, most of the existing methods in …

Content truncated to 500 words