Study On Similarity Analysis Of Event Streams In Real-time Monitored Applications

Event stream management has become a very popular research issue in database field and gained wide attention in recent years. Actually, lots of applications are event-driven. For example, the metadata collected by RFID equipments are simple events. The transaction and business data changes also can be considered as events in businiess process management. In these application areas, these continuous arrival events constitute event streams without bound.Event stream is a special representation form of real-time stream data, which consists of continuous event items. Event stream has characteristics of infiniteness, instantaneousness, real-time, order, and semantic-abundance and so on. Due to all these characteristics and the requirements of applications, the traditional data stream management technique can not satisfy the challenges, and therefore new techniques are required to handle all these problems.In order to detect the complex events, frequent patterns and abnormal events from huge volume of event streams, we need to analyze the relationship between multiple event streams. Hence the similarity analysis of event streams is the key problem of event stream management and knowledge discovery. In this paper, we analyze the similarity relationship between event streams. Our main contributions are as follows:1) The definition of event stream similarity is proposed. The Weighted-Edit-Distance is adopted to measure the similarity, which is computed by the dynamic programming. This measurement can appropriately reflect the similarity of event stream pairs.2) The basic method has high time complexity and do not satisfy the requirement of quick arrival event streams, so we propose a CO-occurrence based algorithm which is more efficient. This method utilizes the simple computed CO-occurrence to quickly filter the unsuitable event streams pairs, reduce the size of candidate set, and accelerate the analysis process.3) The sliding window mechanism is adopted due to the dynamic characteristic of event stream. An effective updating mechanism is proposed for computing of CO-occurrence. In this way we can acquire the higher time efficiency without reducing the precision.4) Similar range of event streams has two characteristics, which are uncertain start position and uncertain length. So similarity analysis at windows of fixed length will lose some results. We design the algorithm to search the local similar regions when the global windows are dissimilar, and we also give the start and end position of local similarity. According to the theoretical analysis and experimental evaluations, it can be proved that the similarity analysis method of event stream have great efficiency and feasibility...