Predicting popularity of online products via collective recommendations.

Predicting the future popularity of commodities has always been a significant issue in information filtering research. Existing methods predominantly rely on the historical popularity of products, assuming that historically popular items will continue to be popular in the future due to preferential attachment. However, this method has limitations as it neglects the intricate structural information within the bipartite networks connecting users and items. The prediction method based on preferential attachment fails for commodities with the same degree of popularity. In this paper, we propose a popularity prediction method that aggregates user recommendation results to forecast item popularity. The method is general and applicable to any recommendation algorithm. For simplicity, we validate the method using the classic collaborative filtering algorithm. Experiments demonstrate that this method significantly outperforms the preferential attachment predictor in accurately predicting the future popularity of niche commodities. [Display omitted] • Preferential attachment fails to work on commodities with the same degree. • The method aggregates recommendations for all users to predict item popularity. • Our method surpasses preferential attachment in niche items popularity prediction. • The method applies to any recommendation algorithm. [ABSTRACT FROM AUTHOR]