"CosMo: Web API Recommendation for Mashup Development via LLM-Guided Collaborative Semantics Modeling"

"A real-world mashup--API dataset constructed from ProgrammableWeb. The dataset contains historical invocation relationships between mashups and Web APIs, together with mashup descriptions, API descriptions, and functional category information. After removing instances with missing attributes, the resulting dataset contains 8{,}217 mashups, 1{,}647 APIs, and 17{,}182 observed mashup--API interactions. Each mashup invokes at least one API, and the average number of invoked APIs per mashup is 2.09. Since the task is to recommend executable APIs from a fixed catalog, we normalize casing and separators to map API names to canonical identifiers.We split mashups into training, validation, and test sets with a ratio of 8:1:1, while retaining all APIs in a shared repository across splits."