Text Classification using Long Short Term Memory

Abstract

Text classification is important for people to navigate and browse through the online document quickly also, for spam filtering, sentiment analysis, news filtering, organizations and many more. Research challenge of text classification is together with a number of classes and number of training examples. As the number of classes increases accuracy will be decreased. This project presents dictionary based approach and LSTM with word2vec for text classification. Dictionary based approach is used for understanding how text classification is done. The main objective of this project is to study the effect of each parameter on LSTM using word2vec and to compare LSTM with other classical methods. In this project, six experiments were conducted on seven different datasets. Datasets are namely IMDB, Amazon review full score, Amazon review polarity, Yelp review polarity, AG news topic classification, Yahoo! Answers topic classification, DBpedia ontology classification. After experimentation, results show that 100 batch size, 50 epochs, Adagrad optimizer, 5 hidden nodes, 100-word vector length, 2 LSTM layers, 0.001 L2 regularization, 0.001 learning rate gives the higher accuracy of LSTM for IMDB, Amazon review full score, Yahoo! Answers topic classification dataset and nearer for Amazon review polarity, Yelp review polarity, AG news topic classification and 6% less for DBpedia ontology classification dataset as compared to be previously the published result of different methods.