Penerapan Natural Language Processing pada Helpdesk IT untuk Otomatisasi Respon Awal

Abstract

IT Helpdesk is a service responsible for handling various information technology problems experienced by users within a company. The initial response process, which still relies on IT staff, can result in suboptimal response times, particularly when the number of tickets increases. This study aims to implement Natural Language Processing (NLP) to automate the initial response to user complaints through the utilization of a Knowledge Base. The study employs the Prototyping method with a microservice architecture, in which Laravel is used as the main application and Flask as the NLP service. The processing stages include text preprocessing, synonym normalization, Term Frequency–Inverse Document Frequency (TF-IDF) weighting, document representation using the Vector Space Model (VSM), and similarity calculation using Cosine Similarity to generate the best solution recommendation (Top-1 Retrieval). In addition, a Confidence Threshold mechanism of 20% is applied to filter recommendations with low similarity scores, allowing tickets to be automatically escalated to the IT Team. Evaluation using 25 test data resulted in Precision, Recall, and F1-Score values of 80.00% each. Robustness Testing also achieved an overall relevance rate of 96%, indicating that the system is capable of handling variations in short queries, synonym usage, and simple typographical errors, although it still has limitations in understanding paraphrases and variations in sentence meaning. The results demonstrate that the implementation of Natural Language Processing based on TF-IDF and Cosine Similarity can support the automation of initial responses in IT Helpdesk services, thereby helping improve ticket handling efficiency.