Combining review elements for modelling various multi-criteria collaborative recommendation models

Traditional single-criterion recommender systems rely on overall ratings, failing to capture accurate user preferences. While multi-criteria recommender systems (MCRSs) address this by leveraging explicit or implicit criteria, existing studies predominantly focus on single review elements, overlooking the potential of combining multiple review elements for richer insights. This paper bridges this gap by proposing novel MCRS models that integrate diverse review elements—such as implicit ratings, aspects, and helpfulness—to enhance recommendation accuracy. A key innovation lies in a novel user profile modelling approach that dynamically combines these elements, enabling granular preference analysis. Comprehensive experiments on the large-scale Amazon dataset demonstrate that the Trust-based Multi-Criteria Similarity with Average Value (TMCSAV) model outperforms all proposed models and the state-of-the-art baselines, achieving the lowest prediction errors (MAE: 0.7473, RMSE: 0.9966) and superior relevance identification (F1-score: 0.65). By prioritising trustworthy users and semantically clustered aspects, TMCSAV mitigates data sparsity and noise, validating the importance of multi-element integration. This work advances MCRS theory through hierarchical aspect clustering and trust-aware paradigms while offering practical value for industries reliant on personalised recommendations, from e-commerce to streaming services.