Long Short Term Memory Convolutional Neural Network for Indonesian Sentiment Analysis towards Touristic Destination Reviews

Abstract

Large amount of text has been created on the Internet which requires assessment to convert this data into useful information. Deep learning can address this challenge by delivering improved performance in sentiment analysis compared to classic machine learning that utilises the statistical technique. LSTM (Long short-term memory), CNN (Convolutional neural network), their combined model, and developments in their architecture have shown excellent performance for assessment of sentiment in English corpus. However, there have been limited research works on deep learning that utilizes a blend of the two models for the Indonesian body of languages. In this research, we present the LSTM-CNN combined model and the Word2Vec framework for assessment of sentiment in the Indonesian language with respect to the reviews of tourist regions. The dataset comprises 10000 touristic destination reviews in the Indonesian language (5000 positive and 5000 negative reviews). The parameters for LSTM-CNN and Word2Vec which were put to test in the study are dropout, pooling layer, learning level, convolutional activation, Word2Vec architecture, Word2Vec evaluation approach, and Word2Vec dimension. The outcomes indicate that the precision of the LSTM-CNN model is higher compared to LSTM; the precision of LSTM-CNN is 97.17% as against 90.82% for LSTM. Going forward, our results could be utilised by the government or the tourism sector as a material basis for fostering tourism, and by the public as a platform for selecting travel destination.