Exploring Alternative Clustering for PIY Source Code Detection

摘要

In this paper, we compare different clustering algorithms given a specific type of data set. Clustering is a powerful tool because it distributes data into meaningful groups based on the information found in data sets. Given a set of data points, each defined by a set of attributes, we find clusters such that points in one cluster are more similar to one another and less similar to points in other clusters. These groups of clusters are crucial to how data is analyzed. It helps us easily identify and give meaning to certain data according to their traits. Clustering helps handle a data set with more utility thus the study of techniques for finding the most representative cluster model is vital in knowledge extraction. Previous work done by Anthony Ohmann and Professor Imad Rahal propose a scalable system called PIY (Program It Yourself) that can detect source code plagiarism over a large repository of submissions where new submissions are compared to current ones. By using clusters, one can compare a new submission to a subset of the data. Accuracy and time are both important factors for PIY. Therefore, we base efficiency of clustering on accuracy and time. In this paper, we perform an analysis of K-Harmonic Means (KHM) against one of PIY's current clustering algorithms called K-Medoid. Developed by Dr. Bin Zhang, the KHM algorithm is derived from the K-Means and Harmonic Average algorithm. It is known to be more "robust" than the K-Means algorithm. Our goal is to find which algorithm gives us the most favorable results.