融合多维信息的主题自适应Web API推荐方法  

Topic-Adaptive Web API Recommendation Method via Integrating Multidimensional Information

作　　者：李鸿超 刘建勋[1] 曹步清[1] 石敏[1] LI Hong-Chao;LIU Jian-Xun;CAO Bu-Qing;SHI Min(Key Laboratory of Knowledge Processing ＆ Networked Manufacturing（Hu＇nan University of Science and Technology）,Xiangtan 411201,China)

机构地区：[1]知识处理与网络化制造湖南省普通高校重点实验室(湖南科技大学),湖南湘潭411201

出　　处：《软件学报》2018年第11期3374-3387,共14页Journal of Software

基　　金：国家自然科学基金(61872139;61873316;61572187);国家科技支撑计划(2015BAF32B01);湖南省自然科学基金(2017JJ2098)~~

摘　　要：如何根据用户的自然语言需求描述自动生成或推荐用于解决问题的WebAPI服务集合,并辅助构建Mashup,是业务流程管理者和服务组合者关注的热点之一.如何提高推荐的质量,是大家关注的焦点.为此,提出了一种融合多维信息的主题自适应Web API推荐方法 HDP-FM(hierarchical Dirichlet processes-factorization machines)为Mashup的创建推荐Web APIs集合.该方法以Web API的描述文档为语料库,利用HDP模型训练每个Web API的主题分布向量;其次,利用已生成的主题模型预测每个Mashup的主题分布向量,用于相似度的计算;最后,将Mashup之间的相似度、WebAPI之间的相似度、Web API的流行度和共现性作为因子分解机模型的输入,评分排序获取用于推荐的WebAPIs集合.为了验证HDP-FM方法的性能,使用从ProgrammableWeb平台上爬取的真实数据进行多组实验,实验结果表明,HDP-FM方法在准确率、召回率、F-measure和NDCG@N等方面具有较好的性能.How to automatically generate or recommend a set of Web APIs for Mashup creation according a user＇s natural language description of requirement is a focus of attention among business process managers and services composition designers. A topic adaptive Web API recommendation method, HDP-FM （hierarchical Dirichlet processes-factorization machine）, is proposed in this paper to recommend a set of Web APIs for Mashup creation. This approach firstly makes the Web API description document as a corpus, and trains a topic distribution vector for a Web API by the HDP model. It then predicts a topic distribution vector for a Mashup via the generated model, where the topic distribution vector is used to calculate the similarity. Finally, a factorization model is utilized to score and sort Web APIs by taking the similarity between Mashups, the similarity between Web APIs, the popularity of Web APIs and the co-occurrence of Web APIs as inputs. A Mashup can be created based on these recommended Web APIs. To verify the performance of the HDP-FM method, a series of experiments are conducted on a real dataset crawled from the ProgrammableWeb platform. The results show that the HDP-FM method has a good performance over others in term of precision, recall, F-measure and NDCG@N.

关 键 词：Web API推荐 HDP(hierarchical DIRICHLET process) 因子分解 Mashup创建 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]