New model of supervised latent semantic indexing
Yixing Liao · Computer Engineering and Applications Journal · 2009
Sprinkling method is a supervised latent semantic indexing integrated with the classification information of the training sample.But the method of feature weight is TF which decreases the text classification performance.And this method doesn’t consider the contribution ability of different samples.In contrast,this method considers the contribution ability of every sample is same.In addition,this method uses several features corresponding to a class label to boost the contribution of class knowledge to classification.A new supervised latent semantic indexing is proposed based on the sprinkling method.The results show that the new model outperforms the sprinkling method.The new model achieves the highest classification performance when feature number is 1,100 which is increased 1.71% compared with the original sprinkling method.