Discovering Factions in the Computational Linguistics Community

Proceedings of the ACL-2012 Special Workshop on Rediscovering 50 Years of Discoveries, pages 22–32, Jeju, Republic of Korea, 10 July 2012. c⃝2012 Association for Computational Linguistics Discovering Factions in the Computational Linguistics Community Yanchuan Sim Noah A. Smith Language Technologies Institute Carnegie Mellon University Pittsburgh, PA 15213, USA {ysim,nasmith}@cs.cmu.edu David A. Smith Department of Computer Science University of Massachusetts Amherst, MA 01003, USA dasmith@cs.umass.edu Abstract We present a joint probabilistic model of who cites whom in computational linguistics, and also of the words they use to do the citing. The model reveals latent factions, or groups of in- dividuals whom we expect to collaborate more closely within their faction, cite within the fac- tion using language distinct from citation out- side the faction, and be largely understandable through the language used when cited from without. We conduct an exploratory data anal- ysis on the ACL Anthology. We extend the model to reveal changes in some authors’ fac- tion memberships over time. 1 Introduction The ACL Anthology presents an excellent dataset for studying both the language and the social con- nections in our evolving research field. Extensive studies using techniques from the field of biblio- metrics have been applied to this dataset (Radev et al., 2009a), quantifying the importance and impact factor of both authors and articles in the commu- nity. Moreover, recent work has leveraged the avail- ability of digitized publications to study trends and influences within the ACL community (Hall et al., 2008; Gerrish and Blei, 2010; Yogatama et al., 2011) and to analyze academic collaborations (Johri et al., 2011). To the best of our knowledge, however, existing work has mainly pursued “macroscopic” investiga- tions of the interaction of authors in collaboration, citation networks, or the textual content of whole papers. We seek to complement these results with a “microscopic” investigation of authors’ interactions by considering the individual sentences authors use to cite each other. In this paper, we present a joint model of who cites whom in computational linguistics, and also of how they do the citing. Central to this model is the idea of factions, or groups of individuals whom we expect to (i) collaborate more closely within their faction, (ii) cite within the faction using language distinct from citation outside the faction, (iii) be largely understandable through the language used when cited from without, and (iv) evolve over time.1 Factions can be thought of as “communities,” which are loosely defined in the literature on networks as subgraphs where internal connections are denser than external ones (Radicchi et al., 2004). The dis- tinction here is that the strength of connections de- pends on a latent language model estimated from ci- tation contexts. This paper is an exploratory data analysis using a Bayesian generative model. We aim both to discover meaningful factions in the ACL community and also to illustrate the use of a probabilistic model for such discovery. As such, we do not present any objective evaluation of the model or make any claims that the factions …

Content truncated to 500 words