ADAMANT — An Anomaly Detection Algorithm for MAintenance and Network Troubleshooting

Network operators are increasingly using analytic applications to improve the performance of their networks. Telecommunications analytical applications typically use SQL and Complex Event Processing (CEP) for data processing, network analysis and troubleshooting. Such approaches are hindered as they require an in-depth knowledge of both the telecommunications domain and telecommunications data structures in order to create the required queries. Valuable information contained in free form text data fields such as “additional_info”, “user_text” or “problem_text” can also be ignored. This work proposes An Anomaly Detection Algorithm for MAintenance and Network Troubleshooting (ADAMANT), a text analytic based network anomaly detection approach. Once telecommunications data records have been indexed, ADAMANT uses distance based outlier detection within sliding windows to detect abnormal terms at configurable time intervals. Traditional approaches focus on a specific type of record and create specific cause and effect rules. With the ADAMANT approach all free form text fields of alarms, logs, etc. are treated as text documents similar to Twitter feeds. All documents within a window represent a snapshot of the network state that is processed by ADAMANT. The ADAMANT approach focuses on text analytics to provide automated analysis without the requirement for SQL/CEP queries. Such an approach provides distinct network insights in comparison to traditional approaches.