Performance Evaluation of Word Embedding Techniques in Sentiment Analysis Based on LSTM Faroh Ladayya (a), Widyanti Rahayu(a), Dania Siregar(a), Ferdiansyah Rizki Saputra(a), Thoriq Akbar Maulana(a)
(a)Faculty of Mathematics and Natural Sciences, Universitas Negeri Jakarta
Abstract
Writing. These opinions can be used as feedback on a product, both goods and services. Sentiment analysis utilized for analyzing opinions given by the public via social media. The sentiment contained in an opinion can be positive, negative or neutral. In this research sentiment analysis makes use of Long Short-Term Memory (LSTM) to classify sentiment into positive, negative or neutral. For classifying sentiment using LSTM, it is necessary to transform text form social media into vector. Thus, word embedding technique are used to convert text into vector. The vector that had been obtained used as input for LSTM. In this study we combine LSTM and several word embedding techniques, namely, Word2Vec, GloVe, ang FastText, Evaluation and comparison were carried out to obtain the best classification performance. Classification performance is compared not only from accuracy, considering that the data used in this research is imbalanced data, it is also important to evaluate the values of precision, recall, F1-score, and ROC/AUC. The analysis results showed that LSTM can effectively learn the sentiment from social media along with word embedding. Further experiments are needed to verify the performance of LSTM with another word embedding technique.
Keywords: Long-Short Term Memory, Sentiment Analysis, Word Embedding
