Using association rule mining to enrich user profiles with research paper recommendation

Abstract

Association rules in recommender systems are used in order to create a model which enriches the users’ profiles but also in addressing cold start problem. Our approach proposes a model, implemented in a system for recommending scientific papers named Collaborative Topic Regression (CTR). Collaborative Topic Regression consists of two matrices, matrix U and V, where matrix U contains the connection between users and paper’s topics, while matrix V contains the connection between papers and the paper’s topics. The CTR system is focused on matrix V, by adopting it, and therefore it is influenced by the textual content of the paper, leaving the content of matrix U almost as it is. The represented model mines association rules from matrix U, and after that, it enriches this matrix using the results from the mining process. The results are based on out-of-matrix prediction and in-matrix prediction. Our approach achieved an increase in quality of the results, in best case for even 20% for the out-of-matrix prediction. Unfortunately, that is not the case also for the in-matrix prediction, which will be further studied in the future.