A data-driven API recommendation approach for service mashup composition

Artikel i vetenskaplig tidskrift, 2025

The increasing availability of Web APIs has brought about a revolution in software development. Developers can now create innovative web applications by combining existing services. However, with so many APIs available, it can be challenging to identify the most suitable ones for a particular task. Many existing recommendation systems rely on keyword matching and historical data, which can limit their effectiveness when dealing with complex functional requirements and new mashup creation scenarios. This paper presents a new method for recommending web APIs to developers for mashup composition. Our goal is to improve the accuracy of recommendations, particularly when developers lack domain knowledge or encounter ambiguous functional descriptions. To achieve this, we propose a solution driven by natural text descriptions, which utilizes advanced techniques such as semantic enrichment and deep learning. The approach to recommendation methods combines content-based and quality-of-service (QoS) techniques with the advanced capabilities of BERT (Bidirectional et al. from Transformers) and Graph Generative Adversarial Networks (Graph GAN). BERT's contextual understanding of text allows us to capture more comprehensive functional descriptions, overcoming the limitations of traditional keyword matching. Meanwhile, Graph GAN helps us learn from existing mashup-service invocation records, leading to more accurate and relevant service recommendations. Our framework consists of a robust data and semantic enrichment component that employs paraphrase mining to extend the vocabulary and enhance semantic similarity measures. As a result, our recommendation system can handle various natural language queries and identify subtle contextual nuances in service descriptions.