Contextual Fine-to-Coarse Distillation for Coarse-grained Response Selection in Open-Domain Conversations

Abstract

We study the problem of coarse-grained response selection in retrieval-based dialogue systems. The problem is equally important with fine-grained response selection, but is less explored in existing literature. In this paper, we propose a Contextual Fine-to-Coarse (CFC) distilled model for coarse-grained response selection in open-domain conversations. In our CFC model, dense representations of query, candidate response and corresponding context is learned based on the multi-tower architecture, and more expressive knowledge learned from the one-tower architecture (fine-grained) is distilled into the multi-tower architecture (coarse-grained) to enhance the performance of the retriever. To evaluate the performance of our proposed model, we construct two new datasets based on the Reddit comments dump and Twitter corpus. Extensive experimental results on the two datasets show that the proposed methods achieve a significant improvement over all evaluation metrics compared with traditional baseline methods.

References in corpus (8)

• Asymmetric LSH (ALSH) for Sublinear Time Maximum Inner Product Search (MIPS)
• Approximate Nearest Neighbor Negative Contrastive Learning for Dense Text Retrieval
• Distilling Dense Representations for Ranking using Tightly-Coupled Teachers
• Is Retriever Merely an Approximator of Reader?
• Speaker-Aware BERT for Multi-Turn Response Selection in Retrieval-Based Chatbots
• Learning an Effective Context-Response Matching Model with Self-Supervised Tasks for Retrieval-based Dialogues
• Dialogue Response Ranking Training with Large-Scale Human Feedback Data
• Ultra-Fast, Low-Storage, Highly Effective Coarse-grained Selection in Retrieval-based Chatbot by Using Deep Semantic Hashing