U-Tapis Melting Words : an Artificial Intelligence Application for Detecting Melt Word Errors in Indonesia Online News

Rapidly advancing NLP technology can assist journalists in correcting linguistic errors, including melt words, which frequently occurs in news writing. A software application called U-Tapis is being developed to assist in correcting spelling mistakes. However, U-Tapis currently lacks the feature to fully correct misspelled words. Therefore, one of the developments being pursued is a Machine Learning model utilizing the Damerau-Levenshtein Distance algorithm and BERT to detect and rectify these errors with high accuracy. Testing of this model yielded F1 Scores of 96.79%, 96.01%, and 96.01% for evaluations on 100, 150, and 200 Tribunnews articles, respectively. The percentages of correct recommendations provided by the model were 88.89%, 85.23%, and 84.68% for the same sets of articles. Therefore, this research is expected to detect errors in word softening and provide appropriate word recommendations, thereby maintaining the quality of language used by journalists and the credibility of a media outlet, which can serve as a model for society.