Who did They Respond to? Conversation Structure Modeling using Masked Hierarchical Transformer

TL;DR

This paper introduces a novel masked hierarchical transformer model that leverages entire conversation history to accurately identify parent utterances, improving conversation structure modeling for applications like summarization.

Contribution

The work proposes a new masking mechanism and transformer architecture that utilize the full ancestral history for better parent utterance prediction in conversations.

Findings

Significant improvement over baselines including BERT on multiple datasets.

Effective modeling of conversation structure using entire ancestral history.

Release of a new large Reddit conversation dataset.

Abstract

Conversation structure is useful for both understanding the nature of conversation dynamics and for providing features for many downstream applications such as summarization of conversations. In this work, we define the problem of conversation structure modeling as identifying the parent utterance(s) to which each utterance in the conversation responds to. Previous work usually took a pair of utterances to decide whether one utterance is the parent of the other. We believe the entire ancestral history is a very important information source to make accurate prediction. Therefore, we design a novel masking mechanism to guide the ancestor flow, and leverage the transformer model to aggregate all ancestors to predict parent utterances. Our experiments are performed on the Reddit dataset (Zhang, Culbertson, and Paritosh 2017) and the Ubuntu IRC dataset (Kummerfeld et al. 2019). In addition, we also report experiments on a new larger corpus from the Reddit platform and release this dataset. We show that the proposed model, that takes into account the ancestral history of the conversation, significantly outperforms several strong baselines including the BERT model on all datasets