基于动态偏好和竞争力的群智化任务推荐

Group-intelligent Task Recommendation Based on Dynamic Preferences and Competitiveness

近年来,群智化软件开发已被学术和工业广泛关注.与传统方法相比,群智化软件开发最大化利用全球开发者资源完成复杂的发展任务,有效降低开发成本并提高发展效率.考虑将推荐技术引入开发者和任务的匹配问题中,即向软件开发者推荐合适的软件开发任务.考虑从两方面解决该问题:一方面,开发人员的任务选择与兴趣偏好的变化有关,因此需要准确捕获.另一方面,软件开发任务与传统商品或其他内容相比具有专业特性,只有相应技能的人才能完成和竞争性质的平台更多,所以开发者也会考虑是否其在众多竞争对手中有较高的评分.因此,研究并完成以下工作:对开发者建模时考虑其动态偏好和竞争力并定义相关的参数指标.提出一个两阶段的群智化软件任务推荐模型:第1阶段使用基于注意力机制的长短期记忆神经网络预测开发者当前的动态偏好,并利用相似度从大量候选任务中筛选出符合偏好的Top-N任务;第2阶段利用开发者的竞争力,使用基于差分进化算法的极端梯度提升方法预测开发者在第1阶段任务上的评分,并按照评分高低向开发者推荐Top-K任务.为了验证其有效性,进行了一系列的实验与已有方法作对比.实验结果表明,所提出的模型在群智化软件任务推荐上有显著优势.

As a novel schema of software development,software crowdsourcing has been widely studied by academia and industry.Compared with traditional software development,software crowdsourcing makes the most use of developers all over the world to complete complex development tasks which can effectively reduce costs and improve efficiency.Nevertheless,because there are a large number of complex tasks in the current crowdsourcing platform and inaccurate task matching will affect the progress and quality of task solutions,it is very important to study the matching problem between developers and tasks.Therefore,this study utilizes the dynamic preferences and competitiveness features of developers and proposes a task recommendation model to recommend appropriate software development tasks for developers.First,the attention mechanism based-long short-term memory network is adopted to predict the current preference of a developer to screen out the top-N tasks that conform to the preference from the candidate tasks.On this basis,according to the developer’s competitiveness,differential evolution algorithm based-extreme gradient boosting is used to predict the developer’s scores of top-N tasks,thus further filtering out the top-K tasks with the highest scores to recommend to the developer.Finally,in order to verify the validity of the proposed model,a series of experiments is carried out to compare the existing methods.The experiment results illustrate that the proposed model has significant advantages in task recommendation in software crowdsourcing.