An enhanced gated recurrent unit with auto-encoder for solving text classification problems

Classification has become an important task for categorizing documents automatically based on their respective groups. Gated Recurrent Unit (GRU) is a type of Recurrent Neural Networks (RNNs), and a deep learning algorithm that contains update gate and reset gate. It is considered as one of the m...

Full description

Bibliographic Details
Main Author: Zulqarnain, Muhammad
Format: Thesis
Language:English
English
English
Published: 2020
Subjects:
Online Access:http://eprints.uthm.edu.my/4928/1/24p%20MUHAMMAD%20ZULQARNAIN.pdf
http://eprints.uthm.edu.my/4928/2/MUHAMMAD%20ZULQARNAIN%20COPYRIGHT%20DECLARATION.pdf
http://eprints.uthm.edu.my/4928/3/MUHAMMAD%20ZULQARNAIN%20WATERMARK.pdf
Description
Summary:Classification has become an important task for categorizing documents automatically based on their respective groups. Gated Recurrent Unit (GRU) is a type of Recurrent Neural Networks (RNNs), and a deep learning algorithm that contains update gate and reset gate. It is considered as one of the most efficient text classification techniques, specifically on sequential datasets. However, GRU suffered from three major issues when it is applied for solving the text classification problems. The first drawback is the failure in data dimensionality reduction, which leads to low quality solution for the classification problems. Secondly, GRU still has difficulty in training procedure due to redundancy between update and reset gates. The reset gate creates complexity and require high processing time. Thirdly, GRU also has a problem with informative features loss in each recurrence during the training phase and high computational cost. The reason behind this failure is due to a random selection of features from datasets (or previous outputs), when applied in its standard form. Therefore, in this research, a new model namely Encoder Simplified GRU (ES-GRU) is proposed to reduce dimension of data using an Auto-Encoder (AE). Accordingly, the reset gate is replaced with an update gate in order to reduce the redundancy and complexity in the standard GRU. Finally, a Batch Normalization method is incorporated in the GRU and AE for improving the performance of the proposed ES-GRU model. The proposed model has been evaluated on seven benchmark text datasets and compared with six baselines well-known multiclass text classification approaches included standard GRU, AE, Long Short Term Memory, Convolutional Neural Network, Support Vector Machine, and Naïve Bayes. Based on various types of performance evaluation parameters, a considerable amount of improvement has been observed in the performance of the proposed model as compared to other standard classification techniques, and showed better effectiveness and efficiency of the developed model.