|
Abstract : |
In this paper, incremental discovery of sequential patterns from a large sequence database is studied. A sequential pattern has the form ff! fi, where ff and fi are subsequences in the database, which says that events ff will be followed by events fi with some minimum support and confidence. We consider the scenario that sequential patterns have previously been discovered and materialized and an update is subsequently made to the database. Rediscovering all patterns by scanning the whole database is not acceptable in a dynamic and large database environment. We propose an incremental discovery algorithm that produces the updated patterns by scanning only the affected part of the database and data structures. In addition, the algorithm handles the dynamism of the minimum support and confidence without recomputation, allowing the user to tune these parameters and focus on most interesting patterns at little overhead. Experiments and comparisons were conducted to test the effectiveness of the proposed algorithm., |
