Semi-supervised Gender Classification with Joint Textual and Social Modeling

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2092–2100, Osaka, Japan, December 11-17 2016. Semi-supervised Gender Classification with Joint Textual and Social Modeling Shoushan Li, Bin Dai, Zhengxian Gong, Guodong Zhou* Natural Language Processing Lab School of Computer Science and Technology, Soochow University, China lishoushan@suda.edu.cn, bdai@stu.suda.edu.cn zhxgong@suda.edu.cn, gdzhou@suda.edu.cn * Abstract In gender classification, labeled data is often limited while unlabeled data is ample. This motivates semi-supervised learning for gender classification to improve the performance by exploring the knowledge in both labeled and unlabeled data. In this paper, we propose a semi-supervised approach to gender classification by leveraging textual features and a specific kind of indirect links among the users which we call “same-interest” links. Specifically, we propose a factor graph, namely Textual and Social Factor Graph (TSFG), to model both the textual and the “same-interest” link information. Empirical studies demonstrate the effectiveness of the proposed approach to semi-supervised gender classification. 1 Introduction Gender classification is a fundamental task with regard to infer user’s gender from the user-generated data. Recently, this task is getting increasingly more attention in some prevailing research fields, such as social network analysis and natural language processing. Applications developed from gender classification have enormous commercial value in personalization, marketing and judicial investigation (Mukherjee and Liu, 2010; Burger et al., 2001; Volkova et al., 2013). In social media, conventional methods handle gender classification as a supervised learning problem over the past decade (Corney et al., 2002; Ciot et al., 2013). In supervised learning approaches, both user-generated textual and user social link features are verified to be effective for gender classification. For instance, in Figure 1, it is easy to infer User c to be a female through analyzing her saying “I'm gonna be a mom!! ” Meanwhile, it is also possible to infer User c is more likely to be a female through analyzing her social link since she follows a cosmetic-selling User “Dior”. Although supervised methods have achieved remarkable success for gender classification, their good performances always depend on a large amount of labeled data, which often need expensive labor costs and long production time. How to learn a classification model with low dependence on the large-scale labeled data becomes an important and challenging problem in gender classification. In this paper, we propose a semi-supervised learning approach to alleviate the above problem in supervised gender classification. Instead of using a large scale of labeled data, we exploit a small scale of labeled data and large amount of unlabeled data to train the model. Our semi-supervised approach employs both user-generated textual knowledge and user social link information. The basic motivation of our approach lies in the observation that social link information might be helpful to infer user gender. Specifically, we focus on the “following” link and think that two users who follow the same particular user could have the same gender. For instance, in Figure 1, User b, User c and User d follow the same user named Dior and they are …

Content truncated to 500 words