Combining Term Weights For Short Text Classification

Text classification is one of the fundamental tasks in the applications of natural language processing(NLP).Convolutional Neural Networks(CNNs)are widely used in NLP tasks.However,applying convolutional neural networks for text classification ignores the discriminating power of words in categories of text classification.Term weighting schemes are widely used in data retrieval and text classification models.Recently,a supervised term weighting scheme was proposed to select those words with high discriminating power of in categories of text classification in a text.To improve CNN for text classification,we introduce a term weighting scheme to improve the CNN for text classification.Here,two separate models were used to perform the task.One is using a term weighting scheme to select words with high discriminating power then use them to represent a text.Another one is using a convolutional neural network to extract features from the new text representation.We examined the single label classification on the dataset Reuters-21578.By using this method,we improved the accuracy of a CNN for text classification.