Topic and Role Discovery in Social Networks

By Andrew McCallum, Andrés Corrada-Emmanuel, Xuerui Wang. International Joint Conference on Artificial Intelligence, 2005

784 citations recorded by the source corpus.

Previous work in social network analysis (SNA) has modeled the existence of links from one entity to another, but not the attributes such as language content or topics on those links. We present the Author-Recipient-Topic (ART) model for social network analysis, which learns topic distributions based on the direction-sensitive messages sent between entities. The model builds on Latent Dirichlet Allocation (LDA) and the Author-Topic (AT) model, adding the key attribute that distribution over topics is conditioned distinctly on both the sender and recipient--steering the discovery of topics according to the relationships between people. We give results on both the Enron email corpus and a researcher's email archive, providing evidence not only that clearly relevant topics are discovered, but that the ART model better predicts people's roles and gives lower perplexity on previously unseen messages. We also present the Role-Author-Recipient-Topic (RART) model, an extension to ART that explicitly represents people's roles.

Subfields: Artificial Intelligence, Statistical and Nonlinear Physics

View Paper: https://www.semanticscholar.org/paper/52a3812faa001fddd3922ac1f6db0c7ad2dea278

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistical and Nonlinear Physics

Citations in the corpus

Topics over time: a non-Markov continuous-time model of topical trends (2006, neutral): “Topics rise and fall in prominence; they split apart; they merge to form new topics; words change their correlations.”
Extracting Social Networks from Literary Fiction (2010, background): “Other recent work in social network construction has explored the use of structured data such as email headers (McCallum et al., 2007) and U.S. Senate bill cosponsorship (Cho and Fowler, 2010).”
A Survey of Statistical Network Models (2009, background): “The related task of discovering the “roles” of individual nodes is useful for identity disambiguation [36] and for business organization analysis [207].”
Topic modeling with network regularization (2008, background): “Although there has been some existing explorations [7, 17, 16, 2], there has not been a unified way to combine textual contents with social networks.”
A Survey of Topic Modeling in Text Mining (2015, background): “The Idea of (ART) is learn topic distributions based on the direction-sensitive messages sent between the senders and receivers [13].”
Topic Models Conditioned on Arbitrary Features with Dirichlet-multinomial Regression (2008, background): “…and McCallum [11] for modeling expertise by multiple topical mixtures associated with each individual author, by McCallum, Corrada-Emmanuel, and Wang [9] for authors and recipients of email, and by Dietz, Bickel and Scheﬀer [4] for inferring the inﬂuence of individual references on citing papers.”
A computational approach to politeness with application to social factors (2013, background): “…of the broader issue of how language relates to power and status, which has been studied in the context of workplace discourse (Bramsen et al., ; Diehl et al., 2007; Peterson et al., 2011; Prabhakaran et al., 2012; Gilbert, 2012; McCallum et al., 2007) and social networking (Scholand et al., 2010).”
Tractable Models for Information Diffusion in Social Networks (2006, background): “Recently, considerable attention has been devoted to investigating social networks [9,5,4,7,11,6], since the progress of the Internet, the World Wide Web, and blogs has enabled us to collect real large-scale social networks.”
Echoes of power: language effects and power differences in social interaction (2011, methodology): “Additionally, researchers have used text features other than linguistic coordination to identify status differences [12, 33, 39, 5]; in contrast with our work these methods picked up situation-specific cues, such as the word “termination” for the Enron corporate-email corpus [12], which are unlikely”
Anonymizing Social Networks (2007, background): “For example, a large corpus of approximately 500,000 email messages, derived from the legal proceedings surrounding the 2001 bankruptcy of the Enron corporation and made public by the Federal Energy Regulatory Commission, has been frequently analyzed by researchers [18].”
Sociology in the Era of Big Data: The Ascent of Forensic Social Science (2015, neutral): “These studies have focused on website linkages (Barabasi 2003), e-mail exchanges (McCallum et al. 2005), and relational data drawn from an assortment of social media, from following ties on Twitter (González-Bailón et al.”
Mining Text Data (2012, background): “This intuition has been studied extensively in work on automatic evaluation of content selection in summarization, where another indicator of divergence—Jensen Shannon divergence—has proven superior to KL [45, 52].”