Chatbot Application “Konco Ngobrol” in Ngapak Javanese Using LSTM Model Approach

Regional languages are an important part of cultural identity, increasingly marginalized in the digital era. Ngapak Javanese, a local dialect, is declining in use in Banyumas, Tegal, Purbalingga, and nearby areas. This research aims to develop an interactive text-based chatbot named Konco Ngobrol that uses Ngapak Javanese as the main conversational language. This chatbot is built using the Long Short-Term Memory (LSTM) model approach, which excels at understanding conversation context in the form of sequential data. The dataset is composed of many sentence patterns with a mixed language style of Indonesian and Javanese Ngapak. The research stages include data collection, pre-processing (tokenization, padding, and label encoding), building and training the LSTM model, model evaluation, prediction using a combination of LSTM and fuzzy matching, and deployment into a Flask-based application with an interactive chat interface. The evaluation results show that the LSTM model can achieve an accuracy of up to 99.04% with a low loss value, and can provide contextually appropriate responses. The fallback system also successfully handled unknown inputs well through fuzzy logic. The Konco Ngobrol chatbot not only serves as a medium for entertainment and communication but also as a tool for preserving digital culture by reintroducing local dialects into the modern technology ecosystem.