Aspect-based Active Learning for User Preference Elicitation in Recommender Systems.
María Hernández-Rubio, Alejandro Bellogín, Iván Cantador · Hispana · 2020
Recommender systems require interactions from users to infer personal preferences about new items. Active learning techniques aim to identify those items that allow eliciting a target user’s preferences more efficiently. Most of the existing techniques base their decisions on properties of the items themselves, for example according to their popularity or in terms of their influence on reducing information variance or entropy within the system. Differently to previous work, in this paper we explore a novel active learning approach focused on opinions about item aspects extracted from user reviews. We thus incorporate textual information so as to decide which items should be considered next in the user preference elicitation process. Experiments on a real-world dataset provide positive results with respect to competitive state of the art methods