Classification of Criticisms and Suggestions on Public Services at RSI Nashrul Ummah Lamongan Using K-Competitive Autoencoder.

Abstract

One form of effort to develop an institution is by listening to feedback from its customers. The institution's response to the feedback received has very important benefits to increase customer satisfaction. In addition, to be able to improve and develop services, data about the customer's perceived condition are needed which will be used for performance evaluation. RSI Nashrul Ummah receives criticism and suggestions centrally so that the input received must be forwarded to the relevant division. The number of inputs is quite large so the information cannot be followed up quickly. The main purpose of this study is to classify incoming complaints and suggestions using the K-Competitive Autoencoder for Text (KATE) method to facilitate the process of redirecting complaints and suggestions. This study modifies the model by adding dropouts, and adjusting the threshold value adaptively so that a good representation model is produced. The testing data amounted to 725 data, able to classify complaints and suggestions with F1-measure in the Administration, Facilities, Personnel, and Service classes above 85%, so that if on average it produces an F1-measure of 87.2%. The F1-measure value is calculated using a multi-label classifier using a confusion matrix for each class.