On recommendation of graph mining algorithms for different data

• Published in: 2016 International Conference on Big Data and Smart Computing (BigComp) pp. 357 - 360
• Main Authors: Tai, Chih-Hua, Lee, Tsung-Han, Chiang, Sheng-Hao, Tsai, Jui-Yi, Yang, De-Nian, Wu, Yi-Hsin, Chan, Ya-Huei
• Format: Conference Proceeding Journal Article
• Language: English
• Published: IEEE 01.01.2016
• Subjects: Algorithm design and analysis; Algorithms; Categories; Classification algorithms; Companies; Conferences; Data mining; Economics; Feature extraction; graph mining; Graphs; pattern mining; recommendation; Social networks; Support vector machines
• ISSN: 2375-9356
• DOI: 10.1109/BIGCOMP.2016.7425947

Inspired by the coming of data-driven innovation and economy, an increasing number of companies over the world are eager to analyze their data for creating useful knowledge, while graph data have become more and more crucial in many areas, such as social networks and medical/chemical applications. Different from conventional transaction data, finding the frequent patterns in a graph is more challenging because graph structures are much more flexible and generalized, and various algorithms have been proposed to properly cope with different graph data. However, for companies and organizations without sophisticated and experienced data scientists, it is usually difficult for them to properly choose a graph mining algorithm that is the most efficient and effective one for their own data. When an inadequate algorithm is employed, excess processing time is usually incurred, and important large patterns may not always be able to be generated. To address the above important need, this paper proposes a new mechanism, referred to as GMRecommend, for recommending a proper graph mining algorithm given the graph data with specific features. GMRecommend is based on support vector machine (SVM) by incorporating two important categories of features: graph features and pattern features. Experimental results manifest that GMRecommend can effectively choose the most proper graph mining algorithm for different kinds of graph data with different characteristics and requirements.