Design And Implementation Of Algorithm Recommendation System For Community Detection

As an important method of graph data analysis and mining,community detection has received great attention from academia and industry during the past decade.Although there are many different community detection algorithms,there is still a lack of an effective system,which can recommend the appropriate algorithm to the graph according to its characteristics.In the meantime,the system should be able for researchers to choose different community detection algorithms to do experiments conveniently.To settle problems above,based on the generalized framework of community detection algorithms,combined with the feature extraction of the graph,the similarity measure and the accuracy measure of the community detection algorithm,this paper proposes and implements anThe primary contributions are summarized as follows:· Design an algorithm recommendation system for community detection which is called CDREC.The system contains multiple functional modules,including the algorithm-execution module,the algorithm-recommendation module,and the vi-sualization module.What's more,CDREC can execute community detection al-gorithms according to users' choice,and the appropriate algorithm can be recom-mended for the users' input datasets while all the execution results can be visually displayed.· Based on the generalized framework for community detection,twelve represen-tative community detection algorithms are implemented to form the community discovery algorithm library.In order to ensure the efficiency of the algorithms,the algorithms are realized by C++.In order to enhance the general applicability of the algorithms,interfaces implemented by C++,Python and Web Service are provided.The algorithm-execution module of CDREC is implemented.· To solve the problem that which community detection algorithm is most suitable for the given graph before the algorithms are executed,this paper proposes an algorithm recommendation model and corresponding algorithm for community detection.For the graph data,feature extraction based on random walk,feature extraction based on 2-hop neighbor structure,and feature extraction based on normalized neighbor structure are proposed.Based on the NMI of the results of the community detection algorithms and the characteristics obtained by different extraction methods,the algorithm recommendation model based on deep convolution neural network is proposed.The experimental results on the LFR synthesis data set show that the algorithm proposed in this work has a good practicality Finally,this paper combine the model and algorithm to form the algorithm-recommendation module of CDREC.· As the visualization of the above results,at the same time to enhance the practicality of this work,this paper implements the visualization module of CDREC.So that the system can show the data structure,community detection results,community evaluation results and community algorithm recommendation results,etc.,for the researchers and ordinary users to provide a good operating experience.