A frequency count approach to multi-criteria recommender system based on criteria weighting using particle swarm optimization

Collaborative filtering (CF) is the most successful and extensively used recommendation technique in the field of Recommender Systems (RS). Generally, the collaborative filtering technique suffers from sparsity and cold start problems because of the absence of appropriate rating information for efficient similarity computation among users. To handle such issues and to identify similar users more efficiently, it is necessary to include other useful user-item details into the system through efficient fusion methods. The multi-criteria information has been proved as a beneficial and influential factor for improving the performance of the classical CF technique. In this research, we incorporate multi-criteria ratings into traditional CF technique using a frequency count approach. The similarity among users is computed using a newly derived common rating weight similarity (CRS) measure. Further, different users show different importance to various criteria of an item. Therefore, a particle swarm optimization algorithm is used to learn optimal weights on different criteria's in the process of global similarity computation. The extensive experimental study driven on a benchmark dataset demonstrates significant improvements in prediction and recommendation qualities compared to the most commonly used similarity and heuristic approaches. (C) 2021 Elsevier B.V. All rights reserved.