Tracking the Dynamic Evolution of Participants Salience in a Discussion

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 313–320 Manchester, August 2008 Tracking the Dynamic Evolution of Participant Salience in a Discussion Ahmed Hassan University of Michigan hassanam@umich.edu Anthony Fader University of Michigan afader@umich.edu Michael H. Crespin University of Georgia crespin@uga.edu Kevin M. Quinn Harvard University kquinn@fsa.harvard.edu Burt L. Monroe Pennsylvania State University burtmonroe@psu.edu Michael Colaresi Michigan State University colaresi@msu.edu Dragomir R. Radev University of Michigan radev@umich.edu Abstract We introduce a technique for analyzing the temporal evolution of the salience of par- ticipants in a discussion. Our method can dynamically track how the relative impor- tance of speakers evolve over time using graph based techniques. Speaker salience is computed based on the eigenvector cen- trality in a graph representation of partici- pants in a discussion. Two participants in a discussion are linked with an edge if they use similar rhetoric. The method is dy- namic in the sense that the graph evolves over time to capture the evolution inher- ent to the participants salience. We used our method to track the salience of mem- bers of the US Senate using data from the US Congressional Record. Our analysis investigated how the salience of speakers changes over time. Our results show that the scores can capture speaker centrality in topics as well as events that result in change of salience or influence among dif- ferent participants. 1 Introduction There are several sources of data that record speeches or participations in debates or discus- sions among a group of speakers or participants. Those include parliamentary records, blogs, and news groups. This data represents a very important and unexploited source of information that con- tains several trends and ideas. In any debate or discussion, there are certain types of persons who c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. influence other people and pass information or ad- vice to them. Those persons are often regarded as experts in the field or simply influential peo- ple and they tend to affect the ideas and rhetoric of other participants. This effect can be tracked down by tracking the similarity between different speeches. We can then imagine a debate with many people arguing about many different things as a network of speeches or participations interacting with each other. We can then try to identify the most salient or important participants by identify- ing the most central speeches in this network and associating them with their speakers. When we have a large dataset of debates and conversations that expand over a long period of time, the salience of participants becomes a dynamic property that changes over time. To capture this dynamic nature of the process, the graph of speeches must evolve over time such that we have a different graph at each instance of time that reflects the interaction of speeches at this instant. We apply our method to the US Congressional Record. The US Congressional Record documents everything said and done in the …

Content truncated to 500 words