Sentiment Analysis in the Mobile Application Review Document Using the Improved K-Nearest Neighbor Method
Indriati Indriati, Ari Kusyanti, Dea Zakia · 2019
The large number of smartphone users makes Indonesia a very potential market for business, especially mobile application developers. Each application store allows application users to provide a review of the applications used. From the existing application store, none has the sentiment analysis feature to filter or categorize positive reviews and negative reviews. Sentiment analysis is computational research of opinions, sentiments, and emotions that are expressed textually. The review document will go through several stages starting from preprocessing, calculation of term weighting to the calculation of cosine similarity (degree of similarity) to the training data used. The next process is sorting the level of similarity, determining the new k-values to produce a category for the document. The testing process of Sentiment Analysis in the Mobile Application Review Document Using the Improved K-Nearest Neighbor Method resulted in the best average accuracy of 88,76%. The best accuracy from all scenario test is 90,67 % and optimal k value is 15. Where the number of documents, comparison or balance of the proportion of training data and the determination of the value of k-values used influence the good or not classification process of the document in the form of an application review.