Annotating Network Service Fault Based on Temporal Interval Relations

The internet has greatly revolutionized the communication and has undoubtedly affects our everyday life from work to entertainment. In order to uphold the quality of network service, Communication Service Providers (CSPs) are striving to keep network service faults to a minimum. To achieve this, they need to detect early of any potential network problems and resolve service incidents promptly before customers are impacted. However, to train a supervised learning algorithm to automatically detect service disruptions, the training data needs to be labeled. It is certainly costly and time consuming process to rely on domain experts to annotate the data. This paper addresses the data annotation problem based on temporal interval relations. We evaluated our method on real-world data and compared it with baseline method.