Data Mining and Knowledge Discovery Handbook, 2 Edition part 113. Knowledge Discovery demonstrates intelligent computing at its best, and is the most desirable and interesting end-product of Information Technology. To be able to discover and to extract knowledge from data is a task that many researchers and practitioners are endeavoring to accomplish. There is a lot of hidden knowledge waiting to be discovered – this is the challenge created by today’s abundance of data. Data Mining and Knowledge Discovery Handbook, 2nd Edition organizes the most current concepts, theories, standards, methodologies, trends, challenges and applications of data mining (DM) and knowledge discovery. | 1100 Zhongfei Mark Zhang and Ruofei Zhang the factoring is two-fold . both regions and images in the database have probabilistic representations with the discovered concepts. Another advantage of the proposed methodology is its capability to reduce the dimensionality. The image similarity comparison is performed in a derived K-dimensional concept space Z instead of in the original M-dimensional code word token space R. Note that typically K M as has been demonstrated in the experiments reported in Section . The derived subspace represents the hidden semantic concepts conveyed by the regions and the images while the noise and all the non-intrinsic information are discarded in the dimensionality reduction which makes the semantic comparison of regions and images more effective and efficient. The coordinates in the concept space for each image as well as for each region are determined by automatic model fitting. The computation requirement in the lower-dimensional concept space is reduced as compared with that required in the original code word space. Algorithm 3 integrates the posterior probability of the discovered concepts with the query expansion and the query vector moving strategy in the code word token space. Consequently the accuracy of the representation of the semantic concepts of a user s query is enhanced in the code word token space which also improves the accuracy of the position obtained for the query image in the concept space. Moreover the constructed negative example neg improves the discriminative power of the probabilistic model. Both the similarity to the modified query representation and the dissimilarity to the constructed negative example in the concept space are employed. Experimental Results We have implemented the approach in a prototype system on a platform of a Pentium IV GHz CPU and 256 MB memory. The interface of the system is shown in Figure . The following reported evaluations are performed on a general-purpose
