SEFR: A Mashup Recommendation Approach for Crossover Service Convergence

Driven by cloud computing, edge computing, and IoT, crossover service convergence, as a key approach for digital economy, aims to generate new services by means of a service composition technique. Mashup services recommendation, by recommending cooperatable application program interfaces (APIs) for the composite service, has emerged as an indispensable technique for crossover service convergence. However, the users’ requirements for mashup applications are complex and diverse. Traditional mashup services recommendation methods face multiple challenges: 1) user intents often have overlapping semantic boundaries that are difficult to distinguish, 2) interaction data between users and services is sparse, and 3) input features come from heterogeneous sources with inconsistent representations. This article proposes semantic-enhanced and feature-fused recommendation (SEFR), a semantic-enhanced and fusion-based mashup services recommendation approach for cross-domain service integration. First, SEFR employs SimCSE (i.e., simple contrastive learning of sentence embeddings), a Transformer-based unsupervised contrastive learning model, to extract discriminative semantic features from user inputs. Second, K-nearest neighbor (KNN) is utilized to retrieve semantically similar instances to enhance current feature representation. Third, an attention gate mechanism is leveraged to adaptively integrate the enhanced heterogeneous features. Finally, SEFR outputs the API list and category prediction results in parallel through sigmoid-activated linear layers, jointly optimized with a supervised contrastive loss. Experimental results show that SEFR improves NDCG@1 and NDCG@10 by 4.51% and 10.67%, respectively, compared to the best baseline.