MavenRank: Identifying Influential Members of the US Senate Using Lexical Centrality

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 658–666, Prague, June 2007. c⃝2007 Association for Computational Linguistics MavenRank: Identifying Influential Members of the US Senate Using Lexical Centrality Anthony Fader University of Michigan afader@umich.edu Dragomir Radev University of Michigan radev@umich.edu Michael H. Crespin The University of Georgia crespin@uga.edu Burt L. Monroe The Pennsylvania State University burtmonroe@psu.edu Kevin M. Quinn Harvard University kevin quinn@harvard.edu Michael Colaresi Michigan State University colaresi@msu.edu Abstract We introduce a technique for identifying the most salient participants in a discussion. Our method, MavenRank is based on lexical cen- trality: a random walk is performed on a graph in which each node is a participant in the discussion and an edge links two partici- pants who use similar rhetoric. As a test, we used MavenRank to identify the most influ- ential members of the US Senate using data from the US Congressional Record and used committee ranking to evaluate the output. Our results show that MavenRank scores are largely driven by committee status in most topics, but can capture speaker centrality in topics where speeches are used to indicate ideological position instead of influence leg- islation. 1 Introduction In a conversation or debate between a group of people, we can think of two remarks as interact- ing if they are both comments on the same topic. For example, if one speaker says “taxes should be lowered to help business,” while another argues “taxes should be raised to support our schools,” the speeches are interacting with each other by describ- ing the same issue. In a debate with many people arguing about many different things, we could imag- ine a large network of speeches interacting with each other in the same way. If we associate each speech in the network with its speaker, we can try to iden- tify the most important people in the debate based on how central their speeches are in the network. To describe this type of centrality, we borrow a term from The Tipping Point (Gladwell, 2002), in which Gladwell describes a certain type of person- ality in a social network called a maven. A maven is a trusted expert in a specific field who influences other people by passing information and advice. In this paper, our goal is to identify authoritative speak- ers who control the spread of ideas within a topic. To do this, we introduce MavenRank, which measures the centrality of speeches as nodes in the type of net- work described in the previous paragraph. Significant research has been done in the area of identifying central nodes in a network. Vari- ous methods exist for measuring centrality, includ- ing degree centrality, closeness, betweenness (Free- man, 1977; Newman, 2003), and eigenvector cen- trality. Eigenvector centrality in particular has been successfully applied to many different types of networks, including hyperlinked web pages (Brin and Page, 1998; Kleinberg, 1998), lexical net- works (Erkan and Radev, 2004; Mihalcea and Ta- rau, 2004; Kurland and …

Content truncated to 500 words