Introduction: This work proposes using feature intensity to improve recommendation accuracy. Despite many proposed recommender systems, to the best of our knowledge, few works have studied the feature intensity factor, which implies that features in items have multiple degrees of strength rather than being binary (exist/not exist). This concept similarly extends to user preferences toward these features. Such intensity can improve the accuracy of the recommendation. Method: In this work, an improved feature intensity-based recommender system is proposed. As part of this work, two novel methods are introduced: an intensity-based User Preference Extractor that extracts the intensity of user preferences from item features, and a Feature Intensity Predictor for Items that predicts the intensity of each feature for a given item. Results: Experimental results demonstrated the effectiveness of the proposed recommender system, with F1 reaching 0.85 for XGBoost in feature-intensity prediction. Furthermore, the proposed system successfully extracted user preferences even in the presence of noise. Discussion and Conclusion: This work showed the impact of incorporating item feature intensities and user preference intensities on improving the accuracy of recommender systems.
Hawashin et al. (Tue,) studied this question.