Inferring Lurkers' Gender by Their Interest Tags

Gender prediction has evoked great research interests due to its potential applications like personalized search, targeted advertisement and recommendation. Most of the existing studies rely on the content texts to build feature vector. However, there is a large number of lurkers in social media who do not post any message. It is unable to extract stylistic or syntactic features for these users as they do not have content information. In this paper, we present a novel framework to infer lurkers’ gender by their interest tags. This task is extremely challenging due to the fact that each user only has a few (usually less than 10) and diverse tags. In order to solve this problem, we first select a few tags and classify them into conceptual classes according to social and psycholinguistic characteristics. Then we enlarge the conceptual class using an association mining based method. Finally, we use the conceptual class to condense the users feature space. We conduct experiments on a real data set extracted from Sina Weibo. Experimental results demonstrate that our proposed approach is quite effective in predicting lurkers’ genders.