A conversational collaborative filtering approach to recommendation

Recent work has shown the value of treating recommendation as a conversation between user and system, which conversational recommenders have done by allowing feedback like “not as expensive as this” on recommendations. This allows a more natural alternative to content-based information access. Our research focuses on creating a viable conversational methodology for collaborative-filtering recommendation which can apply to any kind of information, especially visual. Since collaborative filtering does not have an intrinsic understanding of the items it suggests, i.e. it doesn’t understand the content, it has no obvious mechanism for conversation. Here we develop a means by which a recommender driven purely by collaborative filtering can sustain a conversation with a user and in our evaluation we show that it enables finding multimedia items that the user wants without requiring domain knowledge