摘要
云时代,云API作为服务交付、数据交换和能力复制的最佳载体,已成长为当今面向服务软件开发和企业数字化转型不可或缺的核心要素.然而动态开放网络中持续增长的云API在给开发者提供了更多选择的同时,也将其淹没在海量的云API选择之中,设计有效的云API推荐方法就此成为API经济健康发展中迫切要解决的现实问题.但是,现有研究主要利用搜索关键词、服务质量和调用偏好进行建模,生成质量高功能单一的云API推荐列表,没有考虑服务化软件实际开发中开发者对多元化高阶互补云API的客观需要.高阶互补云API推荐旨在为多个查询云API生成多元互补云API列表,要求推荐结果与查询云API均互补,以满足开发者的联合需求.针对此问题,本文提出基于概率逻辑推理的高阶互补云API推荐方法(Probabilistic Logic Reasoning for High-order Complementary Cloud API Recom⁃mendation,PLR4HCCR).首先,通过云API生态真实数据分析论证云API互补推荐需求的必要性和互补关系建模中替补噪声的客观存在,为云API互补推荐问题研究提供动机和数据支持.其次,采用Beta概率嵌入对云API及其之间的关系约束进行编码,以刻画云API间互补关系的不确定性和支持互补逻辑推理.接着,设计由投影、取反和交并三个基本逻辑算子构建的互补关系逻辑推理网络,使查询集中的每个云API获得非对称互补关系感知和替补噪声消解约束下的互补云API表示.然后,引入注意力机制为查询云API的互补云API分配不同权重,增强高阶互补云API基向量的表征能力.在此基础上,采用KL散度度量高阶互补云API基向量与候选云API之间的距离,并根据KL散度排序生成高阶互补性可感知下的云API推荐结果.最后,我们利用两个真实云API数据集在不同阶互补推荐场景下进行实验,实验表明,与传统启发式推荐方法和深度学习推荐方法相比,PLR4HCCR在互补关系感知推理和替补噪声消解方面均具有较大的优势,继而使其在低阶、高阶和混合阶互补云API推荐中均展示出更优的推荐效果和更强的泛化能力.进一步,超参数敏感性实验、实例分析和用户调查验证了方法的有效性、实用性和可行性,这使结合高阶互补关系的云API推荐方法PLR4HCCR不仅更有可能生成开发者满意的结果,而且可有效提升云API服务提供者的收益.
In the cloud era,cloud API,as the best carrier for service delivery,data exchange,and capability replication,has grown into an indispensable core element in today’s service-oriented software development and enterprise digital transformation.However,while the continued growth of cloud APIs in dynamic open networks provides developers with more options,it also inundates them with a sea of cloud API selection,so designing effective cloud API recommendation methods has become an urgent practical task to be resolved in the healthy development of the cloud API economy.But current existing researches focus on modeling and generating high-quality,single-function cloud API recommendation list using search keyword,service of quality,and call preference,without considering the objective concerns of developers for diversified and high-order complementary cloud APIs in actual service-oriented software development.The purpose of high-order complementary cloud API recommendation is to generate a diversified list of complementary cloud APIs for multiple query cloud APIs,requiring the recommendation results to be complementary to the query cloud APIs to meet the joint interests of developers.To solve this problem,we propose a probabilistic logic reasoning based high-order complementary cloud API recommendation(PLR4HCCR)approach.Firstly,the necessity of cloud API complementary recommendation requirements and the objective existence of substitution noise in complementary relationship modeling were demonstrated through the analysis of real cloud API ecological data,providing motivation and data support for the research on cloud API complementary recommendation problem.Secondly,Beta probability embedding is used to encode cloud APIs and their relationship constraints,so as to characterize the uncertainty representation and support logical reasoning of complementary relationships between cloud APIs.Thirdly,a complementary relationship logical reasoning network is designed,consisting of three basic logical operators:projection,negation,and intersection,so that each cloud API in the query set can obtain the complementary cloud API representation under the constraints of asymmetric complementary relationship perception and substitution noise resolution.Then,an attention mechanism is introduced to assign different weights to the complementary cloud APIs for querying cloud API,enhancing the representation ability of the higher-order complementary cloud API base vector.On this basis,KL divergence is used to measure the distance between the higher�order complementary cloud API base vector and the candidate cloud APIs,and the complementary cloud APIs are generated based on KL divergence ranking under the higher-order complementarity perception.Finally,we conducted a series of experiments using two real cloud API datasets in different order complementary recommendation scenarios.The experiments show that,PLR4HCCR has significant advantages in complementary relationship perception reasoning and substitution noise resolution compared to traditional heuristic recommendation methods and deep learning recommendation methods,which demonstrates better recommendation performance and stronger generalization ability under low-order,high-order,and mixed-order complementary cloud API recommendation scenarios.Furthermore,the hyperparameter sensitivity experiments,case analysis and user study have verified the effectiveness,practicality and feasibility of PLR4HCCR,which makes the cloud API recommendation combined with high-order complementarity not only more likely to generate satisfactory results for developers,but also effectively improve the revenue of cloud API service providers.
作者
陈真
谢登辉
王小龙
孙梦梦
刘啸威
申利民
CHEN Zhen;XIE Deng-Hui;WANG Xiao-Long;SUN Meng-Meng;LIU Xiao-Wei;SHEN Li-Min(School of Information Science and Engineering,Yanshan University,Qinhuangdao,Hebei 066004;The Key Laboratory for Computer Virtual Technology and System Integration of Hebei Province,Yanshan University,Qinhuangdao,Hebei 066004)
出处
《计算机学报》
EI
CAS
CSCD
北大核心
2024年第8期1922-1948,共27页
Chinese Journal of Computers
基金
国家自然科学基金(No.62102348,No.62276226)
河北省自然科学基金(No.F2022203012)
中央引导地方科技发展资金项目(236Z0103G)
河北省教育厅高等学校科技计划项目(No.QN2020183)
河北省创新能力提升计划项目(No.22567626H)资助。
关键词
面向服务软件开发
云API推荐
高阶互补
逻辑推理
Beta概率嵌入
service-oriented software development
cloud API recommendation
high-order complementarity
logical reasoning
Beta probabilistic embedding
