Knowledge-Enriched Graph Convolution Network for Hybrid Explainable Recommendation from Review Texts and Reasoning Path

In the digital era, organizations leverage technology to handle vast datasets, yet deriving meaningful insights remains a challenge. Recommender systems address this by using deep learning advancements to provide users with relevant content. However, ensuring transparency in recommendations, known as explainable recommendations, is difficult due to complex deep learning algorithms. Clarifying recommendations to both users and system designers to enhance comprehension and decision-making represents a significant research challenge in the field of explainable recommendation systems. Review-based explanations in human-like languages are intuitive to the general users while Reasoning-path based explanations are more informative to the system designers. The existing models focus on one of the above explanation approaches. This research proposes a hybrid explainable recommender model integrating review-based and reasoning path-based interpretation. The model utilizes external knowledge bases and review documents to create contextualized knowledge-enriched embeddings for users, items, and interactions. Path encoding with LSTM and attention mechanism is utilized for both rating prediction and reasoning path explanation. A Gated Recurrent Unit with a Copy Mechanism generates human-like recommendation explanations. The model yields both types of recommendation explanations and rating predictions comparable to state-of-the-art technologies.