Research on the categorization accuracy of different similarity measures on Chinese texts

摘要

This paper works on the most intensively studied algorithm- k Nearest Neighbor algorithm. The purpose is to investigate the performance of different similarity measures in the kNN on Chinese texts. The two measures that we focus on are cosine value and Jensen-Shannon Divergence. We use both the corpus collected from the Sogou, whose data extracts from the website of Sohu.com, and datasets that we have processed from real word. The results of our experiment indicate that difference of similarity metrics significantly affects the categorization accuracy.