Opinion Mining-Enhanced Collaborative Filtering for Book Recommendations

Traditional collaborative filtering (CF) approaches in book recommendation systems often suffer from sparsity and limited ability to capture nuanced user preferences. We propose an Opinion Mining-enhanced Collaborative Filtering (OM-CF) model that integrates Aspect-Based Sentiment Analysis (ABSA) into matrix factorization to address these limitations. Specifically, aspect-level sentiments extracted from user reviews are encoded into sentiment vectors and incorporated into the user latent representations. This allows the recommendation model better to reflect users’ detailed opinions toward various book attributes. We evaluate our approach on three benchmark datasets: Goodbooks-10k, Amazon Books, and Book-Crossing. Experimental results demonstrate that OM-CF consistently outperforms strong base-lines, including Matrix Factorization (MF), Sentiment-Aware Recommendation (SAR), and DeepCoNN, across rating prediction and top-N recommendation tasks. These findings highlight the potential of combining opinion mining techniques with collaborative filtering to build more accurate, interpretable, and user-centric recommendation services.