Publications

Abstract : The commercial applicability of the recommendation system (RS) motivates researchers to broaden their focus from beyond accurate recommendations to diverse, novel, and serendipitous recommendations. To improve the capabilities of the personalized recommendation list of the user, we need to view the RS as the multi-objective optimization-based recommendation system. The superiority of the multi-objective recommendation model be subject to address three significant issues: (a) traditional rating evaluation approach has limited competency towards data sparsity issue which in turn affects fitness value, (b) strength of elementary objective functions lack in judging the comprehensive strength of associated component metrics, and (c) relevant recommendations failed to surprise the user with an unexpected yet novel recommendation. Hence, this research proposed a multi-objective recommendation framework (MORF) that jointly optimizes diversity and serendipity metrics with recommendation accuracy. Furthermore, the MORF integrates the reinforced similarity-based implicit similarity computation and rating prediction model to overcome the data sparsity issue in traditional rating metrics. Within MORF, we design three conflicting objective functions to develop the recommendation system’s capability to produce a diverse, surprising, yet relevant recommendation. The generated Pareto front over two benchmark data sets describes the trade-off recommendation solution. Finally, the proposed MORF is evaluated and compared with other baselines in terms of mean precision, diversity, and novelty.