Modification Metric of Class Document on Naïve Bayes for Sentiment Analysis of Online Learning Evaluation

Evaluation of learning in the context of online teaching and learning also has the advantage of flexibility in time and place for students to complete learning evaluations. However, the weakness of online is the large amount of data in Indonesian, making it difficult to make decisions in determining learning outcomes. Another problem is the use of words in Indonesian, which have various positive and negative meanings. This research solves the problem of analysis sentiment in learning evaluation using Artificial intelligence. This research used a modification of metric for features in the naïve Bayes algorithm. Metric feature selection method and naïve Bayes classification to analyze the analysis sentiment of the learning evaluation data, which is done by determining the GU value for each training data feature in the negative and positive classes that have gone through the preprocessing and weighting stages, then feature selection is carried out and is made by combining the two-word classes to determine and sort the GU values from the largest to the lowest. The results obtained are 89.7% accuracy, 89.7% processing, 89.7% recall and 77.5% Fl-Secore. Overall, this method is 45.3% better at predicting the results of learning evaluations in document form than without using feature selection.