EVOLUTIONARY CLUSTERING ALGORITHM DEVELOPMENT FOR COMPLEX HETEROGENEOUS GRAPH

• Main Author: MARYADI - NIM : 23515028 , EDY
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/21879
• Institution: Institut Teknologi Bandung

Graf is often used as a model in many fields, so the resulting graph model is diverse, ranging from simple graphs that model friendships in social media applications known as homogeneous graphs, to complex heterogeneous graphs such as graphs modeling bioinformatics. Processing of complex graphs, especially community detection or cluster formation is still static, ie processing for a complete graph. The processing of a set of graphs with evolutionary clustering is still limited to homogeneous graphs, and heterogeneous graphs with certain schemes. There is no evolutionary clustering algorithm for complex heterogeneous graphs. This thesis will discuss about developing evolutionary clustering method for temporal complex heterogeneous graph. The algorithms GPNRankClus and ENetClus are adopted as algorithms for the cluster process of each complex graph snapshot and evolutionary clustering. By analyzing the evolutionary clustering algorithm, study each stage of the algorithm and perform an analysis of the evolutionary clustering algorithm. Based on the results of the analysis, the development of evolutionary clustering algorithm for complex heterogeneous graphs. The development is done by adding several stages to the GPNRankclus algorithm based on EnetClus evolutionary clustering algorithm. The added stage receives input from the clustering result from the previous snapshot which is the parameter values of the gamma and poisson distributions and auxiliary variable, in addition the previous ranking score is added to the rank score obtained by certain weights. Experiments were performed with some temporal graphs. The experiment results show the evolutionary clustering algorithm for complex heterogeneous graphs gives consistent clustering results between snapshots. The weight for the ranking score affects the consistency between snapshots and the quality of clustering results. Based on the research, it can be concluded that to perform evolutionary clustering on complex heterogeneous graphs can be done by developing clustering graph algorithms that handle complex heterogeneous graphs, by adding stages that receive input from previous results so as to provide consistent clustering results. <br />