Domain Structures in Filtering Irrelevant Frequent Patterns

Events are used to monitor many types of processes in several technical domains. Computers and efficient electronic communication networks make it very easy to increase the accuracy and amount of logged details. While the size of logs is growing, the collection and analysis of them are becoming harder all the time. Frequent episodes offer one possible method to structure and find information hidden in logs. Unfortunately, as events reflecting simultaneous independent processes are stored to central monitoring points, signs of several unrelated phenomena get mixed with each other. This makes the algorithm searching for frequent episodes to produce accidental and irrelevant results. As a solution to this problem, we introduce here a notion of domain constraints that are based on distance measures, which can be defined in terms of domain structure and used taxonomies. We also show how these constraints can be used to prune irrelevant event combinations.