Content-based group recommender systems: A general taxonomy and further improvements

Abstract

Group recommender systems have emerged as a solution to recommend interesting, suitable, and useful items that are consumed socially by groups of people, rather than individually. Such systems have pushed for the use of new recommendation methods within such an emerging scenario, in which the use of the collaborative filtering paradigm is the core of the recommender algorithm. However, collaborative filtering presents several drawbacks and limitations in this scenario, such as the need for lots of rating values, as well as their co-occurrence across several items and users (scarcity). In order to overcome these drawbacks, this research explores a taxonomy for content-based group recommendation systems (CB-GRS), and subsequently the paper discusses and analyzes three specific models that can be used to build CB-GRS, which are (1) CB-GRSs supported by recommendation aggregation and individual ranking, (2) CB-GRSs supported by recommendation aggregation and user-item matching, and (3) CB-GRSs supported by the aggregation of user profiles. Furthermore, the paper presents a hybrid CB-GRS that combines the models (2) and (3) and integrates feature weighting and aggregation function switching. An experimental protocol over well-known datasets is then developed in order to evaluate the proposals. The current study aims at providing a basis to develop a research branch concerning content-based group recommender systems.