KG-ADS: A Log Anomaly Detection Assisted Decision-Making with Support of Knowledge Graph and Reinforcement Learning

There are various methods for log anomaly detection, and different methods have different performances in accuracy, recall rate, and F1 score when they process the different logs. There is often a difficulty in decision-making for log anomaly detection. This paper proposes KG-Ads which is an assisted decision-making with the support of knowledge graph and reinforcement learning , which is designed to reduce the difficulty of applying log anomaly detection. With the support of knowledge graphs, KG-Ads holds a log anomaly detection knowledge which contain 22015 entities, and 11837 relationships, and 118 relationship models. KG-Ads builds an inference engine based on reinforcement learning, and produce recommended policy with the support of the log anomaly detection knowledge base. Our experiments show that KG-Ads is able to provide the suitable policies which are comparable to known policies in the existing work.