The Effect of Embedding Dimension Reduction on Increasing LSTM Performance for Sentiment Analysis

Increasing the amount of data in sentiment analysis research become a problem that requires a different approach. Traditional machine learning has limitations in handling large amounts of data (big data era). Deep learning approach appears to be one of the approaches that can be used. Deep learning in some research can provide good results, especially in NLP research. This research will use the LSTM approach to do sentiment analysis. The input data is modeled into sequence form. This modeling input data into sequence form is expected to improve better accuracy in sentiment classifying. In addition, an analysis of changing in dimension reduction of embedding data with keras framework was experimented to determine its effect on LSTM performance.