Synthetic Dialogue Generation for Interactive Conversational Elicitation & Recommendation (ICER)
This addresses data scarcity for conversational recommender systems, enabling better fine-tuning of language models, though it is incremental in improving simulation methods.
The paper tackled the lack of public data for conversational recommender systems by developing a method to generate synthetic dialogues that are consistent with user states, resulting in a large open-source dataset rated as highly consistent, factual, and natural.
While language models (LMs) offer great potential for conversational recommender systems (CRSs), the paucity of public CRS data makes fine-tuning LMs for CRSs challenging. In response, LMs as user simulators qua data generators can be used to train LM-based CRSs, but often lack behavioral consistency, generating utterance sequences inconsistent with those of any real user. To address this, we develop a methodology for generating natural dialogues that are consistent with a user's underlying state using behavior simulators together with LM-prompting. We illustrate our approach by generating a large, open-source CRS data set with both preference elicitation and example critiquing. Rater evaluation on some of these dialogues shows them to exhibit considerable consistency, factuality and naturalness.