Improving Collaborative Filtering Recommender System Results using Optimization Technique

Nowadays, recommender systems are utilized as a suitable solution to facilitate the shopping process and make it faster. Collaborative Filtering (CF) is the most popular recommendation method which generates the recommendation for the Active User (AU) based on like-minded users. Thus, the selected neighbors have a significant impact on the accuracy and quality of recommendation. This paper presents a novel optimization-based recommender system called Opt-Nibors. OptNibors employs an optimization tool to select the best neighbors' list that improves prediction accuracy. Consequently, each individual represents a candidate neighbors list of AU. The proposed method consists of two phases, preprocessing and optimization phases. The preprocessing phase prepares the used seeds for initializing the population in the optimization phase. The executed preprocessing steps differ based on the historical recorded shopping behavior of AU. A set of experiments was conducted to compare OptNibors with alternative methods. On average, OptNibors improved the prediction accuracy and recommendation quality by 31.1% and 7.7%. The results demonstrate the superiority of OptNibors and its capability to achieve high performance regardless of the number of selected neighbors.