面向Web API挖掘的深度神经网络构建研究

Web API挖掘是API经济时代重要的应用需求，缺乏高效API语义表示、API注释能力不足、面向组合API推荐智能不足是亟待解决的技术挑战。为此，开展面向Web API挖掘的深度神经网络构建研究: (1)针对API关联的多源信息，构建神经网络模型从内容特征、网络特征和多源融合的角度实现API语义深度表示，形成基础组件支持聚类、标注和推荐等应用；(2)针对API注释需求，构建深度神经网络用于注释实体识别与提取，结合协同过滤和API语义表示建立深度注释模型及参数学习方法，优化标注结果；(3)面向API组合需求，构建基于深度神经网络的 API 匹配模型，与建模组合逻辑的生成模型实现联合参数优化，形成基础推荐模型，进一步通过优化候选组合，提升API推荐效果。研发实验系统辅助数据采集和算法测试。项目实施将建立API语义表示、注释增强和组合推荐的深度学习技术框架，为API挖掘研究提供新方法与新思路。

Web API mining is an important application requirement in the era of API economy. The lack of effectively semantic representations of APIs, an insufficient ability on semantic annotations of APIs, a weak intelligence on mashup-oriented recommendations are the key challenges needed to be resolved. Therefore, we propose to develop new approaches for these issues based on deep learning and neural network modeling: (1) In consideration of multi-sourced information associated with APIs, we will construct neural networks to realize deep representations of APIs, from different views of content feature, network feature and information fusion, to support scenarios of clustering, annotations and recommendations; (2) For the semantic annotations of APIs, we will develop a neural network for entity recognition and extraction, establish a deep annotation model and its parameter learning approach in combination with representations of APIs and collaborative filtering, to optimize final results of annotations; (3)For mashup-oriented recommendation, we will build a matching model based on deep neural networks, realize a joint optimization with a generation model for modeling the logic of mashup, then form a fundamental recommendation model, finally design an algorithm to optimize the combination of candidates and improve the effectiveness of recommendations. In addition, an experimental system will be provided to ensure data collection and algorithmic testing. The implementation of the project will establish a technical framework based deep learning for semantic representation, annotation enhancement and mashup-oriented recommendation of APIs and provide new methods and new ideas for mining API resources.