An Optimized Model for Sentiment Classification using Attention Oriented Hybrid Deep Learning Techniques

Abstract

The exponential growth of social reviews of various services has encouraged many researchers to focus on emotion analysis in recent times. The availability of such huge information helps in analyzing the behavior of end-users for improving the QoS. Text categorization is a major language processing research topic that organizes the disorganized text into useful categories. LSTM and CNN models are employed in several natural language processing (NLP) applications for text-based classification and that offer reliable results. CNN models extract top-level features with help of convolutions and maximum pooling-based layers, whereas LSTM based models acquire long-term relationships between text sequences and therefore more suited to text categorization. The combined hybrid method using these popular deep learning-based models is able to memorize categorization, which slows down the training process. In this research, an optimized attention-oriented model combined with BiLSTM with ConvNet is proposed. Model is trained by utilizing two distinct datasets for performance validation of the proposed model. Comparison of the proposed model is performed with other deep learning techniques and the proposed attention-based model has shown a significant performance improvement. The proposed model produces more accuracy in results in comparison to other classic machine learning models.