Analysis On Developer's Behavior In Open-source Projects

With the development and popularization of the Internet,people's online activities are becoming more and more diversified.A huge amount of behavioral data is left behind.With the advancement of big data storage and processing technology,people gradually have the ability to analyze and utilize these data to obtain new knowledge or create new values.In the meantime,the open source community has become one of the mainstream form of software development,where developers make various contributions and collaborate activities.The analysis of the behavior data of developers,is helpful to the management and improvement of the open-source developing process.Due to its importance,there have been many researches focus on the behavior of developers in open-source software development.However,there exists some shortcomings in these studies: 1.The prediction of developer's behavior has always been a hot research field.However,these studies payed little attention to the evaluation of the predictability of behaviors.Whats more,these studies were not attempt to analyze if there are some factors that could affect the prediction of behavior;2.Recent researches have proposed different models in behavior prediction.Nevertheless,developer's behavior may show various patterns,which could not be well fitted all through a single model.At present,there lacks an adaptive prediction method to adept to these various patterns;3.Around the studies on abnormal behavior analysis,most of them are carried out with case studies,rather than improvements on anomaly detection algorithms.This article focuses on developer behavior in Git Hub,brings out new studies in behavior predictability,prediction,and anomaly detection.The outcomes of this article are mainly reflected in the following aspects:1.A predictability evaluation quota based on permutation entropy is proposed.The quota is applied to various time series extracted from developer's behavior data.Furthermore whether factors in open-source development could affect the predictability quota are analyzed.2.A hybrid linear-nonlinear prediction model is applied to different types of developer's behavior sequences correspondingly,due to the overall consideration of the predictability quota and different components of the sequence.Comparing with some single models,our model shows higher prediction accuracy.3.Numerical,trend,and proportional sequences are defined to represent behavior data.On the other hand,point,fragment,and sequential anomaly of developer's behavior are defined.Frequent set mining are used to calculate the distance,so as to obtain the anomaly index.The validity of this method is verified by experiments.All the above aspects provide a technical approach for the analysis of developer's behavior in open-source communities.