A Hybrid Recommendation System for Board Games using Collaborative and Content-based Filtering

With the rapid growth of the board game industry, players often face challenges in discovering games that match their preferences. Collaborative filtering (CF) or content-based filtering (CBF) based recommendation systems are inherently limited when used on their own because of problems like the cold-start problem and personalization issues. We have proposed a hybrid recommendation system that leverages both CF and CBF to create accurate and personalized board game recommendations. Using user ratings, game metadata (mechanics, category, play time), and similarity metrics, we can make predictions about user's preferences. We use the BoardGameGeek (BGG) dataset and compare our recommendations to the fixed CF and CBF approaches. The experimental results show that the hybrid model provides substantially better results than either of the two approaches alone: our evaluated hybrid model reported a RMSE of 0.891, Precision@10 of 0.36, and Recall@10 of 0.29. Given the results, we have shown that not only does utilizing the hybrid approach improve results, but it resolves the cold-start problem; thus, hybrid approaches should be considered in recommendation systems for board games.