[SOUND] Now we get into clustering evaluation, measuring clustering quality. Measuring clustering quality is an important issue just because clustering is unsupervised measure. We want to evaluate the goodness of clustering results by either some internal or external measures. Unfortunately there's no commonly recognized best suitable measure in practice. But there are three categories of measures we call external measures, internal measures and relative measures. For external measures, we can consider they are supervised, employ criteria not inherent to the datasets itself. That means we may have some prior or expert knowledge. For example, some ground truth. Then we can comparing the clustering results against the prior or expert specified knowledge, using certain clustering quality measure. Then the second kinds of measure are called internal measure, which is unsupervised. That means the criteria derived from the data itself. In that case, we will evaluate the goodness of clustering by considering how well the clusters are separated and how compact the clusters are. For example, we can use silhouette coefficient. The third one is a relative measure. That means we can directly compare different class rings using those obtained via different parameter setting for the same algorithm. For example, For the same algorithm, we use different number of clusters. We may generate different clustering results. We may want to compare those results to, for example, we can use silhouette coefficient to see how nicely, by setting a certain number of clusters, the clusters are better separated and you, within each cluster they are more compact so this is a relative measure to comparing different parameter setting. [MUSIC]