Mohammad Faris Azhar, Rizal Setya Perdana, Dian Eka Ratnawati
Sentiment analysis on social media like Twitter is an NLP process that analyzes sentiment based on unstructured text information. NLP incorporates word-level Information Extraction (IE) as an extra feature and as a regular feature. Previous sentiment analysis always removed unnecessary words like stop-words and the embedding process was always word by word. Textual removal cannot be performed conventionally because IE requires every word and symbol inside of text for optimal performance. research before rarely use IE inside the embedding. This research proposes using IE as a feature extraction method for embedding to represent text numerically. This research also compares the proposed feature extraction method with base-line feature extraction methods commonly used in previous studies. Experiments for this research were conducted on the sentiment140 dataset and were evaluated with accuracy, precision, recall and f1 Score. Based on proposed methodology, this research shows us that using IE for feature extraction with stop-word variation connected gives best evaluation performance on a logistic model with F1-score 0.697. © 2025 IEEE.
Brawijaya University, Malang, Indonesia; Brawijaya University, Faculty of Computer Science, Malang, Indonesia
