Graph Based Ticket Classification and Clustering Query Recommendations through Machine Learning

Abstract

 Abstract  This paper aims at exploring the management of incident ticket within the context of IT Service Management by developing a new method based on graph and machine learning. In contrast to modern solutions, traditional approaches are rather basic and utilize simple manual techniques or, at best, Heaviside step function-based algorithms that cannot adequately control the interactions in the data of an incident. In our case, the work flow relies on Resource Description Framework (RDF) graphs to model the relationships and characteristics of incident tickets. We augment the quality of features fed to machine learning models by obtaining these features from such graphs. We use classifiers to facilitate the classification and categorization of tickets on the website while the clustering of similar incidents is facilitated by clustering techniques making recommendations from queries more relevant. Comparing our method to traditional classification approaches, the following factors are considered: accuracy, complexity, scalability, and explainability. Initial experiments indicate a high degree of accuracy in refining classification precision as well a significant improvement in operational efficiency, indicating better reliance on automated and intelligent systems.