Analisis Sentimen terhadap Kalimat Finansial pada FiQA dan The Financial PhraseBank

Abstract

Analisis sentimen atau bisa disebut juga opinion mining merupakan salah satu tugas utama dari Natural Language Processing (NLP) yang merupakan studi komputasi yang mempelajari tentang pendapat seseorang terhadap suatu topik bahasan atau entitas. Analisis dilakukan dengan algoritma machine learning (pembelajaran mesin) Naïve Bayes, Decision Tree, dan K-Nearest Neighbor dengan membagi sentimen ke dalam dua kategori sentimen yaitu sentimen positif dan sentimen negatif. Data analisis diambil dari Financial Opinion Mining and Question Answering (FiQA) dan The Financial PhraseBank yang terdiri dari 4.840 kalimat yang dipilih dari berbagai berita keuangan dan dianotasi oleh 16 annotator berbeda yang berpengalaman dalam domain finansial. Penelitian ini ditujukan untuk mendapatkan hasil analisis sentimen dengan algoritma terbaik melalui perbandingan performa algoritma machine learning Naïve Bayes, Decision Tree, dan K-Nearest Neighbor terhadap kalimat finansial yang disajikan oleh FiQA dan The Financial PhraseBank. Berdasarkan analisis, didapatkan hasil performa dari masing-masing algoritma dengan nilai akurasi algoritma Naïve Bayes sebesar 78,45%; algoritma Decision Tree dengan nilai akurasi sebesar 77,72%; algoritma K-Nearest Neighbor (k=3) dengan nilai akurasi sebesar 41,25%; dan K-Nearest Neighbor (k=5) dengan nilai akurasi sebesar 37,38%. Analisis sentimen dengan algoritma Naive Bayes memiliki performa paling baik dengan nilai akurasi paling tinggi.

Sentiment analysis or can also be called opinion mining is one of the main tasks of Natural Language Processing (NLP) which is a computational study that studies a person's opinion on a topic or entity. The analysis was performed with machine learning algorithms Naïve Bayes, Decision Tree, and K-Nearest Neighbor by dividing sentiment into two categories of sentiment namely positive sentiment and negative sentiment. The analysis data was taken from Financial Opinion Mining and Question Answering (FiQA) and The Financial PhraseBank which consisted of 4,840 sentences selected from various financial news and annotated by 16 different annotators experienced in the financial domain. This research is aimed at obtaining sentiment analysis results with the best algorithms through comparison of the performance of Naïve Bayes, Decision Tree, and K-Nearest Neighbor machine learning algorithms against financial sentences presented by FiQA and The Financial PhraseBank. Based on the analysis, the performance results of each algorithm were obtained with the accuracy value of the Naïve Bayes algorithm of 78,45%; Decision Tree algorithm with an accuracy value of 77,72%; K-Nearest Neighbor algorithm (k=3) with an accuracy value of 41,25%; and K-Nearest Neighbor (k=5) with an accuracy value of 37,38%. Sentiment analysis with the Naive Bayes algorithm (K=5) performs best with the highest accuracy values.