A Semi-Supervised Text Clustering Approach Based on K-Means Algorithm

This paper proposes a semi-supervised text clustering approach based on standard k-means algorithm. Firstly, the semi-supervised text clustering approach transforms textual data into algebraic vectors in a multidimensional space based on VSM (Vector Space Model) and the prior knowledge about the sample set into a set of constraint, which to be represented as a boolean matric. Secondly, the approach makes use of the constraints to supervise the feature weight learning with gradient descent technique and make use the learned feature weight matrix to optimize the Euclidean distance metric, making the distance metric not only reflect the prior constraint relationship of text set, but also discover latent semantic information between texts. Thirdly, applying the k-means clustering algorithm implements text clustering. Providing experimental evidence shows the approach obtains substantial improvements in text clustering compared with k-means clustering algorithm.