Crowdsourcing Software Developer Recommendation Based On Deep Learning

With the rapid development of information technology and software industry,crowdsourc-ing software development has gradually become a research hotspot in the industry.This de-velopment pattern can make full use of the wisdom of the crowd to complete the task more efficiently and with high quality.However,due to the large number of tasks with different needs in the crowdsourcing platform,many newly released tasks waste a lot of time waiting for developers to participate.In addition,the ability of task participants is not guaranteed,result-ing in more unstable quality of software products.At present,researchers solve the matching problem between tasks and developers from the perspective of developer recommendation,so as to ensure the completion quality of software tasks.However,there are still some problems in the existing methods:(1)the existing content-based recommendation methods often use artifi-cial feature extraction methods,which is difficult to extract deeper feature representation?(2)to solving the problem of matching between tasks and developers,most methods only use simple linear weighting on eigenvalues,which makes it difficult to deeply mine the interactive features between tasks and developers,resulting in poor recommendation results?(3)the existing meth-ods pay more attention to recommending the developers who are most capable of completing the task,do not consider whether the developers are interested in participating in the task,and do not evaluate the recommendation results from multiple different dimensions.Based on the above problems,this work studies and completes the following work:(1)By analyzing the data on crowdsourcing platform,this work proposes the task modeling method based on five tuples and the developer modeling method based on the three dimen-sions of ability,enthusiasm and domain knowledge,and defines the corresponding param-eter indicators.(2)In order to recommend interested participants to the task and improve the task participation,considering the task features,this work proposes a multi-label prediction algorithm based on attention mechanism and deep neural network to deeply learn the impact of task features on participants and predict interested developers.On the basis of screening the interested participants,this paper further proposes a developer task score prediction algorithm based on attention mechanism and deep neural network to evaluate the scores of participants,and recommend the top-k developers with the highest scores among the participants interested in the task,so as to complete the task with high quality.(3)This work conducts a series of experiments on the real crowdsourcing platform data.The ex-perimental results show that the proposed method has better recommendation performance than the previous research work.(4)Based on the above research,a developer recommendation system for Topcoder is designed and implemented,which aims to help users quickly find suitable developers for tasks pub-lished by users and improve development efficiency.