Combining Content Information with Collaborative Filtering for Publication Venue Recommendation

Abstract

This paper addresses the problem of academic venue recommendation by developing a hybrid collaborative filtering model that integrates both behavioral and content information. We propose two complementary strategies for incorporating textual content into the collaborative filtering process: enriching the definition of neighborhoods and enhancing the computation of ratings. To evaluate these approaches, we conduct experiments on two document collections, PMSC-UGR and CORD-19, and benchmark them against two state-of-the-art baselines: A publication-based model, which constructs neighborhoods from authors’ venue rating vectors, and a coauthorship-based model, which relies on shared publications to establish similarity. In addition, we explore alternative neighborhood definitions that capture author similarity through textual features, enabling the derivation of latent venue preferences. Experimental results show that integrating content information consistently improves recommendation quality, either by refining neighborhoods or by adjusting ratings. The findings also highlight the importance of adapting the role of textual content to the characteristics of the dataset, as well as the need to investigate richer text representations to mitigate redundancy effects observed when combining content in both components of the model.