Unsupervised Conversation Disentanglement through Co-Training

Abstract

Conversation disentanglement aims to separate intermingled messages into detached sessions, which is a fundamental task in understanding multi-party conversations. Existing work on conversation disentanglement relies heavily upon human-annotated datasets, which are expensive to obtain in practice. In this work, we explore to train a conversation disentanglement model without referencing any human annotations. Our method is built upon a deep co-training algorithm, which consists of two neural networks: a messagepair classifier and a session classifier. The former is responsible for retrieving local relations between two messages while the latter categorizes a message to a session by capturing context-aware information. Both networks are initialized respectively with pseudo data built from an unannotated corpus. During the deep co-training process, we use the session classifier as a reinforcement learning component to learn a session assigning policy by maximizing the local rewards given by the messagepair classifier. For the message-pair classifier, we enrich its training data by retrieving message pairs with high confidence from the disentangled sessions predicted by the session classifier. Experimental results on the large Movie Dialogue Dataset demonstrate that our proposed approach achieves competitive performance compared to the previous supervised methods. Further experiments show that the predicted disentangled conversations can promote the performance on the downstream task of multi-party response selection.

Structural Characterization for Dialogue Disentanglement

“…Disentanglement is worthy of study. Decoupling messages or clustering conversation threads help with screening concerned parts among contexts, therefore it may be naturally required by passage comprehension, and related downstream dialogue tasks (Elsner and Charniak, 2010;Jia et al, 2020;Liu et al, 2021a), such as response selection, question-answering, etc.…”

“…Earlier works mainly depend on feature engineering Elsner and Charniak, 2010;Yu and Joty, 2020), and use well-constructed handcrafted features to train a naive classifier (Elsner and Charniak, 2010) or linear feed-forward network . Recent works are mostly based on two strategies: 1) two-step (Mehri and Carenini, 2017;Yu and Joty, 2020;Li et al, 2020b;Liu et al, 2021a) and 2) end-to-end (Tan et al, 2019;. In terms of the two-step method, the disentanglement task is divided into matching and clustering.…”

“…Inspired by achievements of pre-trained language models (Devlin et al, 2019;Clark et al, 2020;an, 2019), latest work use BERT to contextually encode the dialogue context Li et al, 2020b). Liu et al (2021b) investigates disentanglement from a different perspective. Their end-to-end co-training approach provides a novel unsupervised baseline.…”

Structural Characterization for Dialogue Disentanglement

“…Due to the aforementioned features of dialogue development, there are always multiple ongoing conversation threads developing in a dialogue history simultaneously, which cause troubles for both human and machine to understand dialogue context or further deal with various reading comprehension tasks (Kummerfeld et al, 2019;Elsner and Charniak, 2010;Joty et al, 2019;Shen et al, 2006;Jiang et al, 2018Jiang et al, , 2021. Therefore, to some extent, disentangling context or clustering conversation threads can make an effective prerequisite for downstream tasks on dialogues (Elsner and Charniak, 2010;Liu et al, 2021a;Jia et al, 2020), as which contributes to screening concerned parts for further machine reading comprehension (MRC) applications.…”

“…Earlier works mainly depend on feature engineering (Kummerfeld et al, 2019;Elsner and Charniak, 2010;Yu and Joty, 2020), and use well-constructed handcrafted features to train a naive classifier (Elsner and Charniak, 2010) or linear feed-forward network (Kummerfeld et al, 2019). Recent works mainly based on two strategies: 1) two-step (Mehri and Carenini, 2017;Yu and Joty, 2020;Li et al, 2020c;Liu et al, 2021a) and 2) end-to-end (Tan et al, 2019;. In the two-step method, disentanglement task is divided into matching and clustering, which means firstly matching utterance pairs to detect reply-to relations and then dividing utterances into clusters according to the matching score.…”