Topic Segmentation and Labeling in Asynchronous Conversations

Abstract

Topic segmentation and labeling is often considered a prerequisite for higher-level conversation analysis and has been shown to be useful in many Natural Language Processing (NLP) applications. We present two new corpora of email and blog conversations annotated with topics, and evaluate annotator reliability for the segmentation and labeling tasks in these asynchronous conversations. We propose a complete computational framework for topic segmentation and labeling in asynchronous conversations. Our approach extends state-of-the-art methods by considering a fine-grained structure of an asynchronous conversation, along with other conversational features by applying recent graph-based methods for NLP. For topic segmentation, we propose two novel unsupervised models that exploit the fine-grained conversational structure, and a novel graph-theoretic supervised model that combines lexical, conversational and topic features. For topic labeling, we propose two novel (unsupervised) random walk models that respectively capture conversation specific clues from two different sources: the leading sentences and the fine-grained conversational structure. Empirical evaluation shows that the segmentation and the labeling performed by our best models beat the state-of-the-art, and are highly correlated with human annotations.

ConVis: A Visual Text Analytic System for Exploring Blog Conversations

“…We also compute average and maximum counts for different types of data to better understand what scale the visualization needs to deal with. These values are computed based on a set of 20 Slashdot blogs which comes with human generated topic annotations [JCN13]. Here, the topics and the sentiment are added, since they can be useful for performing almost all of the tasks in Table 1.…”

“…Since the internet revolution and the subsequent rise of social media, an ever‐increasing amount of human conversations are generated in many different modalities. These conversations are primarily asynchronous in nature, where the participants communicate with each other at different times (e.g., emails, blogs, forums, Twitter, Facebook) [JCN13]. Often many people contribute to the discussion, which can quickly become very long with hundreds of comments and replies.…”

“…While asynchronous conversations comprise emails, blogs, microblogs (e.g., Twitter), and messaging; in this paper we focus on the domain of blogs. In fact, blog conversations often have complex thread structure with large number of participants and comments [JCN13], making it a more challenging problem from the visualization perspective. Once we have characterized our domain, we derive a set of design principles, which then guide the visual encoding and interaction techniques of ConVis.…”

Interactive Exploration of Asynchronous Conversations: Applying a User-centered Approach to Design a Visual Text Analytic System

“…Even the interpretations of topics can vary among people according to expertise and the current task in hand. In fact, during topic annotations by human experts, there was considerable disagreement on the number of topics and on the assignment of sentences to topic clusters (Joty et al, 2013b). Depending on user's mental model and current tasks, the topic modeling results may require to be more specific in some cases, and more generic in other cases.…”

Automatic label generation for news comment clusters

“…We test this hypothesis using a baseline that extracts labels from the comment clusters only. We adopt phrase or term as the most suitable linguistic unit to represent labels as evidenced by several previous studies (Mei et al, 2007;Joty et al, 2013;. This paper is organised as follows.…”