Research Of Online Anomaly Detection On Virtual Property Based On Variable-length Sequence

With the rapid development of the Internet over the past decade,the number of Internet Users increased dramatically.And more and more platforms about virtual property appeared on the Internet.However,while we use these platforms to facilitate our work and life,we are facing a variety of potential risk,such as hacking and virtual property being transferred.Therefore,research on online anomaly detection facing to virtual property has very necessary significance.Firstly,from an online detection perspective,we propose a framework about data stream-based online anomaly detection facing to virtual asset.This framework is divided into two modules,online analysis module and offline analysis module.In the former module,mainly,we extract the data stream to get data profile,and then the user's current behavior pattern,which includes the static properties level and the operation sequence level,is gotten form the data profile.Then the current behavior pattern would be matched with behavior patterns in abnormal behavior pattern libraries and normal libraries.In the offline analysis module,periodically,the data of user behavior in database is calculated to dig out the users' normal and abnormal behavior patterns by using variety of pattern generation algorithms.Furthermore,we propose a pattern generation algorithm on the operation sequence level.The core concept of our algorithm is to use variable-length sequence to describe user's complex behaviors.In the training stage,the normal behavior patterns are gotten by digging out the frequent items in the variable-length sequences.Then,we filter out the low-recognizable behavior patterns from the normal behavior patterns by calculating the IDF value of every behavior pattern.In the detection stage,we also use the variable-length sequences to describe users' current behavior patterns.And when we calculate the decision value,a windowed smooth approach is used.Experimental results show that the algorithm we proposed,the variable-length sequence-based online anomaly detection on virtual property,performs well on both the detection accuracy and computational performance.