Long Short-Term Memory and Word Embedding For Sentiment Analysis of User Review
Lia Silviana, Erna Budhiarti Nababan, Muhammad Zarlis · 2023
Accuracy in training sentiment analysis models for large number of review datasets is affected by the correct classification of sentiment labels. Improving the accuracy of sentiment labels, and text representation also affects the performance of sentiment analysis models. Deep learning methods have been widely used to solve various sentiment analysis problems. To improve the performance of deep learning in sentiment analysis, it is necessary to use the right labeling method and good text representation to be used as an embedding layer. This study proposes sentiment labeling using Lexicon and Long Short-Term Memory (LSTM) as well as used FastText as embedding words in sentiment classification. As a corpus, the InSet Lexicon Dictionary is employed for feature extraction. The sentiment data used is the reviews given by users on several applications provided on Google Play. The results showed that the LSTM network using Word-embedded FastText with a dimension of 300 words received a small error value of 0.111 with an accuracy of 95.55% for data labeled based on Lexicon.