Compound Critiquing for Improving Query Refinement on Conversational Recommender System

A conversational recommender system (CRS) is a form of a recommender system to recommend a product through conversational dialogue. We have developed a CRS based on product functional requirements in previous research. This CRS has good interaction capabilities between the system and the user because it can accommodate users unfamiliar with technical features (novice users). However, when user needs are still general, the system will ask questions again to narrow down the user needs (query refinement). A good query refinement process is when one iteration of query refinement can significantly reduce the number of products that match the query. Thus, the interaction process can be more efficient. The problem with functional requirements-based CRS is that interaction based on functional requirements causes the query refinement process to be slower. Based on this problem, we combine query refinement based on functional requirements and technical features to speed up the query refinement process so that user-system interaction is expected to be more efficient. In this study, we focus on applying compound critiquing for technical features-based query refinement involving compound critiquing. We propose the FP-Growth algorithm for generating frequent itemset because it has the fastest execution time than other algorithms such as Apriori and ECLAT.