Association Rule Guided Web API Complementary Function Recommendation for Mashup Creation: An Explainable Perspective

Abstract

As more and more service providers encapsulate and publish their data and services on the internet in the form of Web APIs, the number of Web APIs is ever-increasing. For this reason, Web API recommendation is gaining momentum and has achieved high performance in accuracy. However, few studies paid attention to function recommendation. As an indispensable entity in the mashup-API ecosystem, function is not only the in-facto basis of API taxonomy, but also determines the compatibility and the internal construction pattern of mashup composition. Considering that users adopt multiple function invocations during a development cycle and the number of functions is also increasing, we propose the complementary function recommendation(CFR), a function-to-function problem. To solve the CFR, we regard each mashup as a transaction set for frequent pattern mining and propose an association rule-based complementary function recommendation(ARCFR) system, which provides function recommendation and corresponding probability explanation. Our experiments show that ARCFR can recommend complementary function effectively, and we give two applicable scenarios to demonstrate the practical value of our method in more aspects.