Web Mining and Social Networking: Techniques and ... - tud.ttu.ee

54 3 Algorithms and Techniquesin the dendrogram indicates a merging of two sub-clusters. The value of the horizontal line(i.e., on y-coordinate) is the similarity between the two sub-clusters merged. The bottom layerof the dendrogram indicates that each object is viewed as a singleton cluster. By moving upfrom the bottom layer to the top one, the dendrogram allows us to reconstruct the intermediatemerges that resulted in the resultant hierarchy of clusters. The pseudo code of the HierarchicalAgglomerative Clustering (HAC) algorithm is presented in Algorithm 3.8.Algorithm 3.8: The hierarchical agglomerative clustering algorithmInput: A dataset DOutput: A hierarchy tree of clustersAllocate each object o in D as a single cluster;Let C be the set of the clusters;While |C| >1 doFor all clusters X, Y ∈ C doCompute the between-cluster similarity S(X,Y);endZ=X ∪Y, where S(X,Y) is the minimum;Remove X and Y from C;C=C ∪ Z;end0.10.2Similarity0.40.60.8o1 o2 o3 o4 o5 o6o7 o8 o9 o10 ObjectFig. 3.16. An example dendrogram consists of 10 objectsIn a brief summary, the advantages of the hierarchical clustering are [29]: (1) enable flexibilitywith regard to the level of granularity; (2) ease of dealing with any form of similaritymetric; and (3) applicability to any attribute type. The disadvantages of the hierarchical clusteringare summarized as: (1) vagueness at judging when to terminate; and (2) the fact thatmost hierarchical algorithms do not revisit intermediate clusters with the purpose of their improvement.