Predicting Author Age from Weibo Microblog Posts

Predicting author age from Weibo microblog posts Wanru Zhang, Andrew Caines, Dimitrios Alikaniotis, Paula Buttery Department of Theoretical & Applied Linguistics University of Cambridge wrzhang2010@gmail.com, apc38@cam.ac.uk, da352@cam.ac.uk, pjb48@cam.ac.uk Abstract We report an author profiling study based on Chinese social media texts gleaned from Sina Weibo (新浪微博) in which we attempt to predict the author’s age group based on various linguistic text features mainly relating to non-standard orthography: classical Chinese characters, hashtags, emoticons and kaomoji, homogeneous punctuation and Latin character sequences, and poetic format. We also tracked the use of selected popular Chinese expressions, parts-of-speech and word types. We extracted 100 posts from 100 users in each of four age groups (under-18, 19-29, 30-39, over-40 years) and by clustering users’ posts fifty at a time we trained a maximum entropy classifier to predict author age group to an accuracy of 65.5%. We show which features are associated with younger and older age groups, and make our normalisation resources available to other researchers. Keywords: Weibo, microblog linguistics, text forensics, computational sociolinguistics 1 Introduction The area of digital text forensics continues to inter- est computational linguists, information engineers and web technologists alike. This interest is based on the hypothesis that demographic factors correlate with language use to some discernible extent, a hypothesis that was famously supported by William Labov’s sur- veys in 1960s New York (Labov, 1966), and has contin- ued to gain support from various sociolinguistic works (Trudgill, 2011; Wieling et al., 2014; Hovy, 2015). Previous work in text forensics has for the most part fo- cused on age and/or gender prediction via shallow text features such as word, character and part-of-speech n-grams (Mukherjee and Liu, 2010; Nguyen et al., 2011; Peersman et al., 2011). Here, we attempt to predict the age groups of Chinese microblog authors using such text features, along with features relating to non-standard use of orthography – e.g. emoticons, hashtags, and repeated characters. Our data source is Sina Weibo (新浪微博), the most popular microblogging service in China with 212 mil- lion monthly active users in June 20151. Sina Weibo is one of several Chinese microblogging sites, with weibo actually meaning ‘microblog’ in English; oth- ers include Tencent Weibo, Sohu Weibo, and NetEase Weibo. However, such is Sina Weibo’s preeminence among its competitors that it is now commonly re- ferred to as ‘Weibo’, a practice we follow henceforth. We collected Weibo posts from users in four age groups: under-18, 19 to 29, 30 to 39, and over 40 years old. We extracted linguistic and orthographic features from these posts, trained a maximum entropy classi- fier, and achieved an F-measure of 65.5% by treating fifty posts from each user as a single document. The implication is that, with microblogs being characteris- 1Source: China Internet Watch, accessed 2015-10-12. This compares to Twitter’s MAU count, 316m, in the same period (source: Twitter, accessed 2015-10-12). tically so brief, there is much benefit in agglomerating many posts from a single user, for improved machine learning and classification. 2 Weibo normalisation We collected 40,000 posts …

Content truncated to 500 words