Conversation Kernels: A Flexible Mechanism to Learn Relevant Context for Online Conversation Understanding
Vibhor Agarwal, Arjoo Gupta, Suparna De, Nishanth Sastry
TL;DR
This work tackles the challenge of understanding online conversations where posts are short and context-dependent. It introduces Conversation Kernels, a two-stage approach that first retrieves task-specific conversational context using two kernel families and then uses a RoBERTa-based context-augmented encoder to predict post categories. The method formalizes p(y|x) as a marginalization over context windows, p(y|x) = \sum_{w} p(y|x,w) p(w|x), and demonstrates substantial performance gains over strong baselines and zero-shot LLMs on a large Slashdot dataset across four categories (funny, informative, insightful, interesting). The results indicate the importance of selecting the right surrounding context and show strong generalizability to unseen data, suggesting broad applicability to tree-structured online conversations and potential extensions to other subjective/objective labeling tasks. Overall, Conversation Kernels provide a flexible, generalizable framework for incorporating structured conversational context into downstream understanding tasks.
Abstract
Understanding online conversations has attracted research attention with the growth of social networks and online discussion forums. Content analysis of posts and replies in online conversations is difficult because each individual utterance is usually short and may implicitly refer to other posts within the same conversation. Thus, understanding individual posts requires capturing the conversational context and dependencies between different parts of a conversation tree and then encoding the context dependencies between posts and comments/replies into the language model. To this end, we propose a general-purpose mechanism to discover appropriate conversational context for various aspects about an online post in a conversation, such as whether it is informative, insightful, interesting or funny. Specifically, we design two families of Conversation Kernels, which explore different parts of the neighborhood of a post in the tree representing the conversation and through this, build relevant conversational context that is appropriate for each task being considered. We apply our developed method to conversations crawled from slashdot.org, which allows users to apply highly different labels to posts, such as 'insightful', 'funny', etc., and therefore provides an ideal experimental platform to study whether a framework such as Conversation Kernels is general-purpose and flexible enough to be adapted to disparately different conversation understanding tasks.