A Comparative Study of Neural Network for Text Classification

Abstract

This With the popularity of artificial intelligence in recent years, Natural Language Processing (NLP) technology has also become the focus of research. NLP technology's unique machine translation and text sentiment analysis functions can prevent people from experiencing poor language communication when travelling abroad and help artificial intelligence understand people's language better. This article has made corresponding practice and analysis for the critical requirement of “text classification” in NLP. In the experiment, we used the Internet Movie Database (IMDB) film review forum as the dataset. Recurrent Neural Network (RNN) and the corresponding variants of RNN (Long Short Term Memory (LSTM)) are analyzed and compared from the theoretical aspect. Moreover, we introduced a bidirectional mechanism to optimize RNN and reduce the influence of parameter changes on model training by comparing specific neural network structures. We found the benefits of LSTM in text classification applications compared with RNN and simple neural networks by comparing experiments. Besides, we explored the role of the bidirectional mechanism for RNN. Finally, we create a two-way LSTM model for text classification model and obtain the model training results indicating less overfitting and less loss than other structures.