Enhancing Sentiment Analysis Using Hybrid Deep Learning

The objective of this research is to enhance sentiment analysis of digital currency investors on twitter with hybrid deep learning. By using two deep learning algorithms, which are Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) is called CNN-GRU. In this work, data from twitter textual content in English languages and Thai languages 1,000 samples were divided into positive message 500 samples and negative message 500 samples. Then prepare data before modeling as tokenization by Attacut algorithm, lower case transformation, stemmer and spilt data 70% to training data and 30% to testing data and also compared by adjust parameters to measure the classification efficiency. The experiment results showed that CNN-GRU was the best performance in the classification of positive message and negative message with Word Embedding Dimension 64, Number of Kernel 128 and Kernel Size 5, Memory 16, and recurrent dropout 0.4. The best result of CNN-GRU was accuracy 82.67%, precision 0.84, recall 0.80, f-measure 0.83, and ROC 0.82.