Conversational Recommender Systems (CRS) aim to provide tailored recommendation responses via a chat interface, including both the user's preferred item and its accompanying explanation. However, due to its generative nature, CRS are prone to responding with factually incorrect explanations (i.e., hallucinations). To solve this problem, we propose incorporating a passage retrieval module into CRS with the objective of enhancing the factuality and informativeness of system responses. Specifically, we outline essential directions for employing a passage retrieval module in CRS to address the following critical issues: (1) the risk of passage retrieval not aligning with the user preference; (2) the absence of supervision for training a passage retrieval module. As a solution, we introduce ESPRESSO, a novel passage retrieval approach for CRS, to effectively tackle the above issues with two core ideas: adaptive item selection and relevance-based groupwise learning. Our extensive experiments show that ESPRESSO effectively resolves issues, achieving up to 36% higher Hit@3 accuracy than the best of 8 competing methods. Additionally, we verify that leveraging passages retrieved by ESPRESSO significantly improves the response quality of CRS.

For high-quality conversational recommender systems (CRS), it is important to recommend the suitable items by capturing the items' features mentioned in the dialog and to explain the appropriate ones among the various features of the recommended item. We argue that the CRS model should be a domain-expert who is (1) knowledgeable about the relationships between items and their various features and (2) able to explain the recommended item with its features relevant to dialog context. To this end, we propose a novel framework, named as LATTE, to pre-train each core module in CRS (i.e., the recommendation and the conversation module) through abundant external data. For the recommendation module, we pre-train the recommendation module to comprehensively understand the relationships between items and their various features by leveraging both multi-reviews and a knowledge graph. For pre-training the conversation module, we create the synthetic dialogs, which contain responses providing the explanation relevant to the dialog context by using all the items' features and dialog templates. Through extensive experiments on two public CRS datasets, we demonstrate that LATTE exhibits (1) the effectiveness of each module in LATTE, (2) the superiority over 7 state-of-the art methods, and (3) the interpretations based on visualization.