Gated Hybrid Collaborative Filtering for Top-N Recommendation

A new recommender system framework, Gated Hybrid Collaborative Filtering (GHCF), integrates textual reviews into an autoencoder-based collaborative model to improve top-N recommendation ranking. The architecture uses an adaptive gating mechanism to balance collaborative embeddings and topic-based features layer-wise. A contrastive learning module aligns semantic and collaborative signals in the latent space. The framework is evaluated across five configurations: pure collaborative, topic and gated, text and gated, and with contrastive learning. The approach addresses the misalignment between review-aware models optimized for rating prediction and the need for discriminative ranking in top-N scenarios.