Improving Recommendation Diversity by Classifying Users Based on Demographic Information

Today, with the explosive growth of the internet applications and in the current age of information overload, recommender systems are steadily becoming more important in filtering relevant information and items for users. So far, there have been many studies on developing new algorithms that can improve the accuracy of recommendations. In contrast, in recent years, several researchers have indicated that it is not sufficient to have accuracy as the sole criteria in measuring recommendation quality and considering other important dimensions such as diversity is necessary to generate recommendations that are not only accurate but also useful to users. However, in most work done in the field of recommendation diversification, it is expressed that the gains of recommendation diversity is often accompanied by the losses of accuracy, the phenomenon called accuracy-diversity dilemma. This paper proposes a classification approach which applies demographic information in order to improve recommendation diversity. Experiments on MovieLens data set show that the proposed approach, while offering significant improvements in recommendation diversity, does not decrease the accuracy and overcomes the accuracy–diversity dilemma. This paper, in addition to its proposed approach, presents a comprehensive review of the work done in the field of recommendation diversification and provides a classification of existing approaches.