Convolutional Neural Network Based Sentence Classification Algorithm

Sentence classification is an important task of Natural Language Process.Recently, neu-ral network based sentence classification algorithms are proved to be effective for sentence sentiment analysis、question classification and other sentence classification tasks as they could use more context information in the models.The major neural networks used in sentence classification are Recursive neural networks, Recurrent neural networks and Convolutional neural networks.We propose two novel sentence classification algorithms called Pre-Trained Word Embed-ding Sentence Algorithm and Dynamic Expanded Sentence Algorithm.(i)Pre-Trained Word Embedding Sentence Algorithm uses the generic distributed word embeddings trained by large corpus, to initialize the input sentences of convolutional neural network.This means the convolutional neural network would contain more general semantic information.(ii) Dynamic Expanded Sentence Algorithm is based on Pre-Trained Word Embedding Sentence Algo-rithm.It would dynamically expand the training sentences those contain less semantic infor-mation by feature selection based on the statistics of training data.The expanded sentences would be used as the input of the convolutional neural network training process.The empirical study on data sets shows that the proposed algorithms are more effective.