Enhancing Dialogue Systems with Discourse-Level Understanding Using Deep Canonical Correlation Analysis

TL;DR

The paper tackles the problem of sustaining coherence and relevance in long dialogues by introducing a discourse-level understanding framework based on Deep Canonical Correlation Analysis (DCCA). By learning a shared latent space between an utterance and its discourse history, the model derives discourse tokens that emphasize salient contextual relationships while filtering irrelevant information. The architecture comprises an encoder, a discourse-level understanding module, and a response selection module, and it demonstrates significant improvements on the Ubuntu Dialogue Corpus across automatic metrics. This approach enhances long-range context management for dialogue systems, with practical implications for applications like customer support and virtual assistants, and sets the stage for further work on speaker intent and sentiment integration.

Abstract

The evolution of conversational agents has been driven by the need for more contextually aware systems that can effectively manage dialogue over extended interactions. To address the limitations of existing models in capturing and utilizing long-term conversational history, we propose a novel framework that integrates Deep Canonical Correlation Analysis (DCCA) for discourse-level understanding. This framework learns discourse tokens to capture relationships between utterances and their surrounding context, enabling a better understanding of long-term dependencies. Experiments on the Ubuntu Dialogue Corpus demonstrate significant enhancement in response selection, based on the improved automatic evaluation metric scores. The results highlight the potential of DCCA in improving dialogue systems by allowing them to filter out irrelevant context and retain critical discourse information for more accurate response retrieval.