Document Modeling with Hierarchical Deep Learning Approach for Sentiment Classification
Monalisa Ghosh, Goutam Sanyal · 2018
Sentiment analysis has recently been considered as most active research field in NLP domain. Deep learning is a growing trend of machine learning due to its automatic learning capability with impressive results across different NLP task. In this paper a model is proposed to analyze the deep sentiment representation based on CNN and LSTM (modified version of RNN) network. We aim to improve the performance of traditional machine learning method by merging them with deep learning techniques to tackle the challenge of sentiment prediction of massive amount of unsupervised product review dataset. We make our model first learn to sentence representation with CNN. Next, the semantics of sentences are encoded with LSTM network for document representation. We conduct experiments on two review datasets based on movie review with evaluation metric 'accuracy'. The result shows that proposed model outperformed traditional machine learning as well as baseline neural network model