Dynamic diversity-enhanced similarity retrieval system and method

摘要

A conversational recommender system retrieves k cases (3) and generates a structure for user feedback (4). The structure includes a list of critique units for each of which the user can specify a value range. It also generates compound critiques, each being a combination of features and value ranges presented together. An explanation of a compound critique can be generated in response to a user request. The system determines (7), from one cycle of a session to the next, if a preference has been carried. If so, a re-focus function implementing diversity is used for retrieval in the next cycle. If not, a refine function implementary similarity-based retrieval is used for the next cycle of the session.