A Keyword-driven Recommender System With Compatible APIs

With the development and popularization of the Internet of Things(IoT),the ever-increasing number of enterprises and organizations encapsulate their business processes into web APIs that can be accessed remotely.On the one hand,the emergence of these lightweight web APIs expands the reputation and influence of API providers.On the other hand,it brings convenience to potential mashup developers exactly.Reusing existing web APIs allows developers to build mashups quickly,which can not only reduce the development periods and costs,but also ensure high-quality mashup development.Although the fast-growing web APIs and their functional diversity in web communities bring more choices for mashup developers,they also bring many difficulties and challenges to developers' APIs selection decision.In light of the above-mentioned difficulties,the APIs recommendation system performs on demand,which can return a set of optimal APIs for developers by analyzing user preferences to alleviate the burden of decision-making.However,the existing APIs recommendation algorithms still have the following two major challenges:(1)Recommendation results are not compatible enough.Existing recommendation algorithms often recommend combinations of APIs according to user preferences,and usually ignore the compatibility between the recommended web APIs.A combination of incompatible APIs may cause the built mashup not stable enough;In worse cases,the mashup fails to work.(2)Recommendation results can't meet the individual needs of developers.In practice,developers generally have exact functional expectations for the mashup to be built.However,existing APIs recommendation methods often fail to insight into the users' development needs accurately in advance,which makes it hard to guarantee that the recommended APIs can fulfill the functions expected by the mashup developers.In this situation,how to recommend a set of APIs that meet the developers' personalized functional requirements and have the best compatibility is becoming one of the key criteria for judging whether the APIs recommendation is successful.Therefore,we propose a novel web APIs recommendation approach named K-CAR(Keywords-based and Compatibility-aware web APIs Recommendation),and apply it to design a personalized-driven and compatible APIs recommendation system.Generally,the major contributions of this article are two-fold.(1)We propose a novel data-driven approach K-CAR for economically and efficiently building mashups.Our approach models the keywords-based and compatibility-aware APIs recommendation problem as a minimal group Steiner Tree problem to search the global optimal solution step by step based on Dynamic Programming(DP).We deploy large-scale experiments on real-world dataset to evaluate the performance of our proposal in terms of multiple criteria.Experiment results demonstrate the usefulness,effectiveness and efficiency of our proposal.(2)We apply the above-mentioned K-CAR to our keyword-driven and compatibility-aware web APIs recommendation system.On the basis of feasibility analysis and demand analysis,we employ SQLite to design the system database,and utilize the Python programming language as well as the Django framework to implement the functional modules of our system,which provides mashup developers with practical and effective APIs compositions.