一种考虑兴趣偏好的Top-k众包开发者推荐方法

A Top-k crowdsourcing developer recommendation method considering interest preference

随着云开发技术的不断发展,以众包软件平台为代表的在线开发者社区成为软件工程领域一个新的研究热点。如何为平台中的任务发布者及时、准确地推荐开发者是具有现实意义的重要问题。本研究提出一种考虑兴趣偏好的Top-k众包开发者推荐方法,改变传统Top-k推荐列表的生成模式,有针对性地为任务推荐符合条件的开发者。首先结合平台丰富的异构信息提取任务特征,考虑兴趣感知特征和评分数据构建开发者初始推荐列表,挖掘列表内开发者多个维度特征,并利用加权融合策略生成开发者综合能力特征;随后通过开发者列表分类器,完成任务与开发者列表的分类匹配,生成候选开发者推荐列表;最后基于候选列表内开发者的相似情况,为任务推荐匹配度最高的开发者列表。此外,提出两种冷启动解决方案,有效缓解推荐时面临的开发者冷启动问题。为了评估整个模型的性能,从ZhuBaJie平台和TopCoder平台爬取数据进行了实验,结果表明,本方法在准确率和覆盖率等多个评价指标上都取得了较好的结果。

The research on online developer communities,which are represented by the crowdsourcing software Platforms,has become a new hot topic in the field of software engineering,with the progress of the times and the development of the Cloud-based services.Then,it is an important issue with practical significance that how to recommend developers for task publishers timely and accurately on this platform.In this paper,a Top-k crowdsourcing developer recommendation method considering interest preference will be proposed,which changes the generation mode of the original Top-k recommendation list,and selectively recommends the qualified developers for the task.Firstly,building a task feature model which is based on the rich heterogeneous information on the platform.Then,the initial recommendation list of developers is constructed considering the preferences of interest perception and rating data,mining multiple dimensional features of developers in the list,and using a weighted fusion strategy to generate a developer comprehensive capability model.After that,completing the classification and matching of different tasks and developer lists by the developer list classifier to generate a candidate list.Finally,recommending the most suitable developer list for the task based on the similarity of developers in the candidate list.In addition,two cold start solutions are proposed,which effectively alleviate the cold start problem of developers.In order to evaluate the performance of this entire model,the extensive experiments have been performed with data crawled from platforms named ZhuBaJie and TopCoder,the results show that the method proposed by this paper has achieved good results on many evaluation indicators such as accuracy and coverage.