An Enhanced Content-based Filtering Using Maximal Marginal Relevance

Purpose–The studyaims toenhance Content-based Filtering by diversifying its recommended items to combatoverspecialization. It traditionallyrecommends items that are directly related to the user profile, preventingusers from discovering newer sets ofitems. Method–Maximal Marginal Relevance isintegrated into the algorithm –are-ranking algorithm,developed by Carbonell and Goldsteinthat enhancesthe diversity of items retrieved by information retrieval systems–to enhance Content-based Filtering and address the underlying overspecialization problem. Results–By integrating Maximal Marginal Relevance, themodified algorithmaddressed overspecialization.Out of all the tested values of lambda (λ) for MMR, theenhanced Content-based Filtering (CBF-MMR)with λ = 0.7showed the most prominence, having a good balance between relevance and diversity of recommendations. Onaverage,itimprovedupon the original algorithm by 48.51% in Precision, 6.40%in Recall, 28.12%inF-Score,and 275.45% in Diversity. Conclusion–Resultsshow that integratingMaximal Marginal Relevance to Content-based Filtering (CBF-MMR) improves the diversity of recommendations. Due to the re-ranking process added by the Maximal Marginal Relevance, the average Precision, Recall, and F-Score also improved. Recommendations–The authors of this studysuggest further workon Content-based Filtering withfasterre-ranking algorithms, application ofthe enhanced algorithm to other larger datasets such as GroupLens’ MovieLens 10M dataset, application of the enhanced algorithm to a different domain,andenhancement of the Maximal Marginal Relevance algorithm to be applied in Content-based Filtering. Research Implications–The successful integration of Maximal Marginal Relevance (MMR) in a Content-based Filtering algorithm opens new possibilities for enhancing the diversity and relevance of recommendations of various types of recommender systems. Practical Implications–Beyond the movie recommender system this study was applied to, thisstudyhas profoundpractical implications on other domains that utilize recommender systems including but not limited to the domains of entertainment, e-commerce, and information retrieval platforms. Keywords–recommendersystem, content-based filtering, maximal marginal relevance, overspecialization