Online shopping is one way that is currently in great demand by the public, especially in Indonesia. By shopping online, especially at Lazada stores, consumers don't need to spend a lot of time and energy. Because of the ease of technology that can now be used in shopping online. But to find out the quality of a product, consumers will see reviews of items that have been sold. Therefore with the number of consumers who write a lot of data collected so that a way is needed to be able to sort out positive or negative sentiments by doing word repairs because of the many word writing errors that we often encounter on a review. So it needs word repairs so that consumers can understand more clearly the contents of a review. In this study the researchers made the system using the Jaro Winkler Distance method which was used to improve the word and then performed scoring calculations with BM25, as well as the classification with K-Nearest Neighbor (KNN). Based on the test results get the best accuracy value of 89% with the value of F-Measure 88% in the second k-fold test with a value of k = 11. So the use of word normalization on training data and improvement of words in the test data can increase the results of sufficient accuracy better than without using word repairs and without normalizing training data.
Copyrights © 2019