Skill Inference with Personal and Skill Connections

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 520–529, Dublin, Ireland, August 23-29 2014. Skill Inference with Personal and Skill Connections Zhongqing Wang†, Shoushan Li†∗, Hanxiao Shi‡, and Guodong Zhou† † Natural Language Processing Lab, School of Computer Science and Technology, Soochow University, China ‡ School of Computer Science and Information Engineering, Zhejiang Gongshang University, China {wangzq.antony, shoushan.li}@gmail.com hxshi@mail.zjgsu.edu.cn, gdzhou@suda.edu.cn Abstract Personal skill information on social media is at the core of many interesting applications. In this paper, we propose a factor graph based approach to automatically infer skills from per- sonal profile incorporated with both personal and skill connections. We first extract personal connections with similar academic and business background (e.g. co-major, co-university, and co-corporation). We then extract skill connections between skills from the same person. To well integrate various kinds of connections, we propose a joint prediction factor graph (JPFG) model to collectively infer personal skills with help of personal connection factor, skill connection fac- tor, besides the normal textual attributes. Evaluation on a large-scale dataset from LinkedIn.com validates the effectiveness of our approach. 1 Introduction With the large amount of user-generated content (UGC) published online every day in the context of social networks (Tan et al., 2011; Luo et al., 2013), such online social networks (e.g., Twitter, Facebook, and LinkedIn) have significantly enlarged our social circles and much affected our everyday life. One popular and important type of UGC is the personal profile, where people post their detailed information, such as education, experience and other personal information, on online portals. Social websites like Facebook.com and LinkedIn.com have created a viable business as profile portals, with the popularity and success largely attributed to their comprehensive personal profiles. Obviously, online personal profiles can help people connect with others of similar backgrounds and provide valuable resources for businesses, especially for personnel resource managers to find talents (Yang et al., 2011a; Guy et al., 2010). In the profiles, the personal skill information is the most impor- tant aspect to reflect the expertise of a person. However, few social platforms allow users to manually attach such personal skill information into their personal profiles. For example, in our collected dataset, 91.8% skills appear less than 10 times. Even the distribution of the top 10 frequently occurring skills is asymmetric, and only 43.1% people attach skills on their profiles. For this regard, it is highly desirable to develop reliable methods to automatically infer personal skills for personal profiles. Although it is straightforward to recast skill inference as a standard text classification problem, i.e., predicting the skills with the profile text alone, personal profiles usually are poorly organized, even with critical information missing. Thus, it is challenging to infer skills given the limited information from the profile texts. We propose two assumptions to address above challenges by incorporating additional connection information between persons and skills: • People are always connected to others with similar academic and business backgrounds (e.g. co- major, co-corporation). For example if there is co-major, co-university, …

Content truncated to 500 words