Web api recommendation for mashup development has become increasingly challenging due to the rapid growth of available apis. Current approaches face two critical limitations: semantic inconsistency in service descript...
详细信息
Web api recommendation for mashup development has become increasingly challenging due to the rapid growth of available apis. Current approaches face two critical limitations: semantic inconsistency in service descriptions and insufficient exploitation of categorical relationships. To address these challenges, we propose SRCA (LLM-enhanced service Semantic Representation and categoryco-occurrence feature Augmentation), a novel framework that integrates LLMs' sophisticated semantic processing capabilities with graph-based categoryco-occurrence relationship modeling and feature enhancement for Web api recommendation. At its core, SRCA employs specially designed requirement-focused and functional-oriented prompts to guide LLMs in transforming diverse service descriptions into unified structures, while leveraging the LLMs' comprehensive semantic understanding capabilities to extract accurate functional features. This semantic understanding is further enhanced by a feature augmentation model that constructs a mashup-api category co-occurrence graph to discover complex service relationships by capturing both explicit categorical similarities and implicit functional correlations while preserving essential semantic characteristics. Extensive experiments on a large-scale dataset from ProgrammableWeb, comprising 7739 mashups and 1342 Web apis across 443 categories, demonstrate SRCA's superior performance over state-of-the-art baselines, with significant improvements in recommendation accuracy (13.89% in Precision@5, 16.96% in MAP@5, and 12.48% in NDCG@5).
暂无评论