Abstract: As a simple and effective classification algorithm, kNN algorithm is widely used in text classification. However, the k value (usually fixed) is usually set by users. For this purpose, the reconstruction and locality preserving projections (LPP) technology is introduced into the nearest neighbor classification, which makes the selection of the k value to be determined by the correlation between the samples and the topology structure. The algorithm uses l₁-norm sparse coding method to reconstruct the test sample by its k (not fixed) nearest neighbor samples and LPP keeps the local structure of the sample after the reconstruction, which not only solves the problem of choosing k value, but also avoids the influence of fixed k value on classification. Experimental results show that the classification performance of the proposed method is better than that of the classical kNN algorithm.

Key words: k-nearest neighbor, locality preserving projections, reconstruction, sparse coding