Research On Open Source Project Team Expansion Via Attributed Network Representation Learning

With the rapid development of the open source community in recent years,the number of open source projects and developers has increased dramatically,and the scale is enormous.How to recommend suitable developers for open source projects to ensure their sustainability has become an urgent problem.The essence of this problem is the matching problem between open source projects and developers.Several works have been done to model the features of projects and developers using deep learning techniques and recommend developers for open source projects by ranking the similarity of their features.These works have shortcomings in modeling the features of projects and developers,which affect the accuracy of their recommendations,mainly in the following aspects: 1)modeling the features of developers focused on their own attributes(e.g.,project experiences,behavioral characteristics,etc.),and pays less attention to the social relationship between developers;2)modeling the features of open source projects focused on the descriptive information about open source projects and pays less attention to the relationship between participating developers and the relationship between projects and their teams.To address these shortcomings,this paper first proposes a cross-fusion based developer attributed network representation learning model for developer feature modeling that considers developers' own attributes and social relationships among developers and then proposes a team expansion recommendation model based on a dual attention mechanism,which effectively models the variability of existing developers' contributions to open source projects and the mutual relationships among developers and the open source project description information.Finally,a prototype team expansion recommender system for GitHub is designed and implemented based on the above techniques.Specifically,the main contributions of this paper are as follows:· Developer attributed network representation learning model via cross-fusion.Network topological information and node attributions are handled in two independent views in this model.Different social groups of each node are partitioned through ego-network partitioning and information are exchanged between views with crossfusion operation based on self-attention mechanism.Finally information is fused into unified embeddings with view weighting operation.Experiments demonstrate that proposed model significantly outperforms existing methods in terms of node classification,node clustering and visualization.· Open source project expansion recommendation model based on dual attention mechanism.Projects embeddings and developers embeddings which are learned with developer attributed network representation learning model via cross-fusion are utilized by this model.Then,mutual relationships between project descriptive information and its team,as well as mutual relationships between team members are modeled by leveraging dual attention mechanism.Finally,developers are recommended according to their matching score with projects.Experiments show that proposed model significantly outperforms existing works in terms of recommendation accuracy.· Team expansion recommender system for GitHub.A prototype recommender system for GitHub whose front end is implemented with React and back end is mainly implemented with Fast API to verify the effectiveness of above techniques.