Characterizing the Language of Online Communities and its Relation to Community Reception

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1030–1035, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Characterizing the Language of Online Communities and its Relation to Community Reception Trang Tran and Mari Ostendorf Electrical Engineering, University of Washington, Seattle, WA {ttmt001,ostendor}@uw.edu Abstract This work investigates style and topic aspects of language in online communities: looking at both utility as an identifier of the commu- nity and correlation with community reception of content. Style is characterized using a hy- brid word and part-of-speech tag n-gram lan- guage model, while topic is represented using Latent Dirichlet Allocation. Experiments with several Reddit forums show that style is a bet- ter indicator of community identity than topic, even for communities organized around spe- cific topics. Further, there is a positive cor- relation between the community reception to a contribution and the style similarity to that community, but not so for topic similarity. 1 Introduction Online discussion forums provide a rich source of data for studying people’s language usage patterns. Discussion platforms take on various forms: arti- cles on many news sites have a comment section, many websites are dedicated to question answer- ing (www.quora.com), and other platforms let users share personal stories, news, and random dis- coveries (www.reddit.com). Like their offline counterparts, online communities are often com- prised of people with similar interests and opinions. Online communication, however, differs from in- person communication in an interesting aspect: ex- plicit and quantifiable feedback. Many discussion forums give their users the ability to upvote and/or downvote content posted by another user. These ex- plicit reward/penalty labels provide valuable infor- mation on the reaction of users in a community. In this work, we take advantage of the available user response to explore the relationship between com- munity reception and the degree of stylistic/topical coherence to such communities. Using hybrid n- grams and Latent Dirichlet Allocation (LDA) topic models to represent style and topic for a series of classification tasks, we confirm that there exists a community language style, which is not simply char- acterized by the topics that online communities are organized around. Moreover, we show that language style is better at discriminating communities, espe- cially between different communities that happen to discuss similar issues. In addition, we found a posi- tive, statistically significant, correlation between the community feedback to comments and their style, but interestingly not with their topic. Finally, we analyze community language on the user level and show that more successful users (in terms of positive community reception) tend to be more specialized; in other words, analogous to offline communities, it is rare for a person to be an expert in multiple areas. 2 Related Work It is well known that conversation partners become more linguistically similar to each other as their dia- logue evolves, via many aspects such as lexical, syn- tactic, as well as acoustic characteristics (Niederhof- fer and Pennebaker, 2002; Levitan et al., 2011). This pattern is observed even when the conversation is …

Content truncated to 500 words