Data quality in recommender systems: the impact of completeness of item content data on prediction accuracy of recommender systems

Recommender systems strive to guide users, especially in the field of e-commerce, to their individually best choice when a large number of alternatives is available. In general, literature suggests that the quality of data which a recommender system is based on may have important impact on recommendation quality. In this paper, we focus on the data quality dimension completeness of item content data (i.e., features of items and their feature values) and investigate its impact on the prediction accuracy of recommender systems. In particular, we examine the increase in completeness per item, per user and per feature as moderators for this impact. To this end, we present a theoretical model based on the literature and derive ten hypotheses. We test these hypotheses on two real-world data sets, one from two leading web portals for restaurant reviews and another one from a movie review portal. The results strongly support that, in general, the prediction accuracy is positively influenced by increased completeness. However, the results also reveal, contrary to existing literature, that among others increasing completeness by adding features which differ significantly from already existing features (i.e., a high diversity) does not positively influence the prediction accuracy of recommender systems.