SATR-API: A SENTIMENT- AND ATTENTION-BASED TRUST-AWARE RECOMMENDER FOR API SERVICES VIA CROSS-MODAL EMBEDDING INTEGRATION

Abstract

In this paper, we introduce SATR-API, a novel hybrid recommendation framework that leverages sentiment cues, trust relationships, and deep contextual fusion for personalized API service suggestions. The proposed system initiates by extracting contextualized sentiment representations from user-generated reviews using a fine-tuned BERT variant. To model user reliability, an adaptive trust network is constructed and embedded using Graph Attention Networks (GAT), capturing personalized trust dynamics more effectively than traditional GCNs. These sentiment and trust signals are integrated into a contextual matrix factorization module enhanced with auxiliary user and service attributes. A cross-modal attention mechanism is employed to unify heterogeneous features, allowing for adaptive feature weighting during the recommendation process. Experiments conducted on a real-world API dataset demonstrate that SATR-API consistently outperforms contemporary approaches in terms of RMSE, MAE, and NDCG. Additional ablation studies highlight the contributions of each component and verify the robustness of the framework.