FSSOM: One novel SOM clustering algorithm based on feature selection

In order to reduce dimension number of feature space and improve clustering precision, a novel SOM clustering algorithm based on feature selection-FSSOM is provided in this paper. This algorithm first evaluates importance and distinguishing ability of each feature, and only selects features which can efficiently improve clustering precision to construct feature space. Then, it computes kullback-leibler divergence of different co-occurring feature vector, which is gotten from large scale training corpus, to reflect the similarity of different feature. This algorithm considers the influences of similar features and uses it in self-organizing-mapping algorithm. It can make latently similar documents into same cluster. The experiment results demonstrate that because of adjusting the similar features' weights, enlarging feature adjusting range, it can efficiently improve clustering precision and reduce training time.