ABSTRACT
Widely spread, recommender systems might face some challenges
such as overspecialization and lack of diversity. In this paper, we
propose a book context-aware recommender system (CARS) that
uses individual characteristics as model features and active search
as a pre-filtering context method in an attempt to increase user’s
newness perception and diversity. To achieve this goal, we revised
literary critic essays to create five binary base-questions able to
separate and aggregate novels through subjective concepts.We also
conducted a data collection to form a dataset around 50 selected
books, evaluated by the public using these questions. Going further,
we developed two recommender systems (RS) using different
strategies to handle imbalanced samples (SELC and SMOTE) and
compare their performance to conclude that SELC generates better
recommendations on an inner performance.