Towards Retrieval-based Conversational Recommendation

TL;DR

This paper explores retrieval-based methods for conversational recommendation systems, demonstrating through a user study that such approaches can produce higher quality responses than recent generation-based models, and highlights their potential as an alternative or complement.

Contribution

The paper introduces a novel retrieval and ranking technique for conversational recommendation systems and provides empirical evidence of its effectiveness through a user study.

Findings

Retrieval-based responses were rated higher in quality than generation-based ones.

The quality ranking of generation models in literature did not align with user study results.

Retrieval approaches can serve as a viable alternative or complement to generation models.

Abstract

Conversational recommender systems have attracted immense attention recently. The most recent approaches rely on neural models trained on recorded dialogs between humans, implementing an end-to-end learning process. These systems are commonly designed to generate responses given the user's utterances in natural language. One main challenge is that these generated responses both have to be appropriate for the given dialog context and must be grammatically and semantically correct. An alternative to such generation-based approaches is to retrieve responses from pre-recorded dialog data and to adapt them if needed. Such retrieval-based approaches were successfully explored in the context of general conversational systems, but have received limited attention in recent years for CRS. In this work, we re-assess the potential of such approaches and design and evaluate a novel technique for response retrieval and ranking. A user study (N=90) revealed that the responses by our system were on average of higher quality than those of two recent generation-based systems. We furthermore found that the quality ranking of the two generation-based approaches is not aligned with the results from the literature, which points to open methodological questions. Overall, our research underlines that retrieval-based approaches should be considered an alternative or complement to language generation approaches.