Memetic Algorithm based Similarity Metric for Recommender System

Abstract

Recommender Systems (RS) are web-based intelligent decision-making tools, which narrow down the user's choices based on their defined and undefined behavior. An evolutionary algorithm, namely, Genetic Algorithm (GA) has shown significant results in the field of RS in the past. Despite its huge success, it suffers from the limitation of premature convergence. Memetic Algorithm (MA), also called parallel or hybrid GA is one such technique which introduces local search to reduce the likelihood of premature convergence. This work presents a novel MA-based Similarity Metric (MASM) for RS, leveraging the collaborative behavior of memes. We use publicly available Movielens dataset (100K ratings) to conduct experiments. Results demonstrate that the proposed metric outperforms the conventional GA-based Similarity Metric (GASM). The precision of RS using MASM is improved by 28% over RS using GASM, resulting in improved predictive recommendation accuracy.