Adaptive trust-aware collaborative filtering for cold start recommendation

• Published in: Behaviormetrika Vol. 50; no. 2; pp. 541 - 562
• Main Authors: Zarei, Mohammad Reza, Moosavi, Mohammad R., Elahi, Mehdi
• Format: Journal Article
• Language: English
• Published: Tokyo Springer Japan 01.07.2023; Springer Nature B.V
• Subjects: Accuracy; Chemistry and Earth Sciences; Cold; Collaboration; Computer Science; Economics; Finance; Insurance; Invited Paper; Management; Mathematics and Statistics; Physics; Ratings & rankings; Recommender systems; Similarity measures; Social networks; Statistical Theory and Methods; Statistics; Statistics for Business; Statistics for Engineering; Trust; Trust propagation; Collaborative filtering; Cold start problem; Recommender systems; Optimization
• ISSN: 0385-7417; 1349-6964
• DOI: 10.1007/s41237-022-00161-3

The most popular techniques in Recommender Systems are based on Collaborative Filtering (CF) approaches. Classical CF learns from similarities among the rating data to predict the missing ratings in generating recommendations. While user ratings are very informative signals, other sources of user data can contribute to the prediction of user preferences. In this paper, we propose a novel adaptive CF approach that can learn from trust connections and social ties. In fact, it can generate high-quality recommendations using side information. The proposed approach is evaluated using the two large datasets Epinions and FilmTrust. The results have shown the effectiveness of our proposed approach compared to several recommendation techniques. Our approach achieves a Mean Absolute Error (MAE) of 0.833 and 0.671 on Epinions and FilmTrust datasets, respectively, in Cold Start situation when no or few ratings have been provided by users to items.