A Comparative Analysis of Long Short-Term Memory Recurrent Neural Network Classifiers for Text Classification

Abstract

 The classification of news text is hard due to tremendous variations in interpretations of text and events such as various groups and vast volume of text contents. Text recognition and classification are challenging research areas in artificial intelligence and machine learning. Deep learning is a type of machine learning that is widely used to build the recognition method for human behavior and text classification systems. A popular deep learning technique Recurrent Neural Network is used to identify the points of interest and remove the attributes from the text for the identification and classification of the human language. The Recurrent Neural Network is a form of neural network in which the output from the previous stage is fed to the current phase data. All inputs and outputs are independent of each other in standard neural networks. It is important to determine next term in a sentence in some situations; the prior words are needed to recall the previous words. Hence, Recurrent Neural Network came in with the aid of a hidden layer, which solved this problem.  The hidden state is the primary and most significant function of Recurrent Neural Network. The Recurrent Neural Network is based on classifiers that use combination of layers. Long Short-Term Memory, Bi-Directional Long Short-Term Memory, Deeper Long Short-Term Memory and Deep Bi Directional Long Short-Term Memory are all Recurrent Neural Network layered base classifiers. They store the previous data block as well as the new data block and merge it into one row as output. In this research, these classifiers are analyzed for a better, generalize and efficient Long Short-Term Memory Recurrent Neural Network text classification classifier. For this purpose, the BBC news platform is selected for text classification. Experiments are conducted on mentioned classifiers and the simulation was done through MATLAB. The comparative findings are presented in form of tables and are analyzed for exploring a better classifier. It is also observed that different evaluation parameters had an effect on the process of text analysis. Deep Bi Directional Long Short-Term Memory has proven its efficiency and better performance comparatively for text classification on performance evaluation measures like Accuracy (74%), True Positive rate (90%), and False Positive rate (6%), Recall (90%), Precision (75%) and F-measure (74%).

    Keywords: machine learning, deep learning, recurrent neural network, convolution neural network, long short-term memory.