Popularity Bias in Correlation Graph based API Recommendation for Mashup Creation

Chao Yan,Weiyi Zhong,Dengshuai Zhai,Arif Ali Khan,Wenwen Gong,Yanwei Xu,Baogui Xin
DOI: https://doi.org/10.1145/3654445
IF: 5
2024-04-02
ACM Transactions on Intelligent Systems and Technology
Abstract:The explosive growth of the API economy in recent years has led to a dramatic increase in available APIs. Mashup development, a dominant approach for creating data-centric applications based on APIs, has experienced a surge in popularity. However, the vast array of choices poses a challenge for mashup developers when selecting appropriate API compositions to meet specific business requirements. Correlation graph-based recommendation approaches have been designed to assist developers in discovering related and compatible API compositions for mashup creation. Unfortunately, these approaches often suffer from popularity bias issues, leading to an inequality in API usage and potential disruptions to the entire API ecosystem. To address these challenges, our research begins with a theoretical analysis of the popularity bias introduced by correlation graph-based API recommendation approaches. Subsequently, we empirically validate the presence of popularity bias in API recommendations through a data-driven study. Finally, we introduce the p opularity b ias aware w eb A PI r ecommendation ( PB-WAR ) approach to mitigate popularity bias in correlation graph-based API recommendations. Experimental results over a real world dataset demonstrate that PB-WAR offers the optimal trade-off between accuracy and debiasing performance compared to other competitive methods.
computer science, information systems, artificial intelligence
What problem does this paper attempt to address?