Abstract:

In order to save the problems of dimensionality curse, noise data in text categorization, the text categorization algorithm was presented based on the non-linear dimensionality reduction algorithm and combined with kNN(knearest neighbor algorithm). The algorithm first removed the noise data, and then used the locally linear embedding algorithm of non-linear manifold learning to recover low-dimensional manifold structure in high-dimensional data to implement dimensionality reduction. The processed data was used to construct k-NN classifiers. Experimental results showed that this  algorithm could  effectively improve the accuracy of text classification.

Key words: data reduction, categorization, non-linear dimensionality reduction algorithm, k-NN