Streaming Analysis of Discourse Participants

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 48–58, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Streaming Analysis of Discourse Participants Benjamin Van Durme Human Language Technology Center of Excellence Johns Hopkins University Abstract Inferring attributes of discourse participants has been treated as a batch-processing task: data such as all tweets from a given author are gathered in bulk, processed, analyzed for a particular feature, then reported as a result of academic interest. Given the sources and scale of material used in these efforts, along with potential use cases of such analytic tools, discourse analysis should be reconsidered as a streaming challenge. We show that un- der certain common formulations, the batch- processing analytic framework can be decom- posed into a sequential series of updates, us- ing as an example the task of gender classifi- cation. Once in a streaming framework, and motivated by large data sets generated by so- cial media services, we present novel results in approximate counting, showing its applicabil- ity to space efficient streaming classification. 1 Introduction The rapid growth in social media has led to an equally rapid growth in the desire to mine it for use- ful information: the content of public discussions, such as found in tweets, or in posts to online forums, can support a variety of data mining tasks. Infer- ring the underlying properties of those that engage with these platforms, the discourse participants, has become an active topic of research: predicting indi- vidual attributes such as age, gender, and political preferences (Rao et al., 2010), or relationships be- tween communicants, such as organizational domi- nance (Diehl et al., 2007). This research can bene- fit areas such as: (A) commercial applications, e.g., improved models for advertising placement, or de- tecting fraudulent or otherwise unhelpful product re- views (Jindal and Liu, 2008; Ott et al., 2011); and (B) in enhanced models of civic discourse, e.g., in- expensive, large-scale, passive polling of popular opinion (O’Connor et al., 2010). Classification with streaming data has usually been taken in the computational linguistics commu- nity to mean individual decisions made on items that are presented over time. For example: assigning a label to each newly posted product review as to whether it contains positive or negative sentiment, or whether the latest tweet signals a novel topic that should be tagged for tracking (Petrovic et al., 2010). Here we consider a distinct form of stream-based classification: we wish to assign, then dynamically update, labels on discourse participants based on their associated streaming communications. For in- stance, rather than classifying individual reviews as to their sentiment polarity, we might wish to classify the underlying author as to whether they are gen- uine or paid-advertising, and then update that deci- sion as they continue to post new reviews. As the scale of social media continues to grow, we desire that our model be aggressively space efficient, which precludes a naive solution of storing the …

Content truncated to 500 words