May 27, 2024

Sajad Ahmadian

Academic rank: Assistant professor
Education: Ph.D in Computer Engineering
Phone: 09188339565
Faculty: Faculty of Information Technology


A novel approach based on multi-view reliability measures to alleviate data sparsity in recommender systems
Type Article
Recommender systems, Collaborative filtering, Reliability measure, Data sparsity, Rating profile
Researchers Sajad Ahmadian، Mohsen Afsharchi، Majid Meghdadi


Recommender systems are intelligent programs to suggest relevant contents to users according to their interests which are widely expressed as numerical ratings. Collaborative filtering is an important type of recommender systems which has established itself as the principal means of recommending items. However, collaborative filtering suffers from two important problems including cold start and data sparsity. These problems make it difficult to accurately compute user similarity and hence to find reliable similar users. To deal with these problems, a novel recommender method is proposed in this paper which is based on three different views of reliability measures. For the first view, a user-based reliability measure is proposed to evaluate the performance of users’ rating profiles in predicting unseen items. Then, a novel mechanism is proposed to enhance the rating profiles with low quality by adding a number of reliable ratings. To this end, an item-based reliability measure is proposed as the second view of the reliability measures and then a number of items with highest reliability values are selected to add into the target rating profile. Then, similarity values between users and also initial ratings of unseen items are calculated using the enhanced users’ rating profiles. Finally, a rating-based reliability measure is used as the third view of the reliability measures to evaluate the initial predicted ratings and a novel mechanism is proposed to recalculate unreliable predicted ratings. Experimental results using four well-known datasets indicate that the proposed method significantly outperforms other recommender methods.