Journal of Artificial Intelligence and Digital Health

ABSTRACT

Efficiently identifying similar textual information is a critical task in modern information technology service desk systems, where ticket volume and redundancy can impede timely responses. In this study, we propose Hybrid SE-FAISS, a deep learning–based framework that combines Semantic Embedding (SE) using a sentence transformer with Facebook Artificial Intelligence Similarity Search (FAISS) for fast and accurate detection of similar textual data. Service desk tickets are transformed into dense numerical vectors using a deep learning model to capture semantic relationships between tickets. These embeddings are indexed using FAISS, a high-performance similarity search technique, enabling rapid retrieval of the most semantically similar text data. The framework was evaluated on a real-world dataset of IT service tickets, including approximately 85,000 entries from healthcare project management systems provided by Probel Yazılım ve Bilişim Sistemleri AŞ, a software company in İzmir, Turkey. The model achieved 80% accuracy in detecting similar tickets based on cosine similarity between embedding vectors. Hybrid SE-FAISS demonstrates that integrating deep learning embeddings with optimized similarity search algorithms provides an effective and scalable solution for text similarity detection, with potential applications in ticket deduplication, automated response recommendation, and knowledge management systems. By rapidly identifying tickets similar to new inputs, the framework reduces response times and enables faster processing through reusable solutions.

Keywords: Artificial Intelligence; Machine Learning; Deep Learning; Text Similarity Detection; Semantic Embedding; Sentence Transformer; Facebook Artificial Intelligence Similarity Search; Service Desk Tickets; Information Technology Helpdesk