DFI
Abstract
Sequential pattern mining, also known as frequency based string mining, has been the objective of various algorithmic approaches over the last years. A recent breakthrough, the linear time algorithm by Fischer, Heun, and Kramer (FHK), has turned out to be theoretically optimal and also quite fast in practice. In 2008, we presented a conceptually much simpler algorithm based on a deferred data structure that is faster and uses less memory at the same time. In this paper, we give an in-depth presentation of this algorithm and show how to use it on multiple databases with a variety of frequency constraints. As such, we use the notion of entropy from information theory to devise the Entropy Substring Mining Problem which is a multiple database generalization of the Emerging Substring Mining Problem. In addition, we evaluate the algorithm rigorously using various string domains, e.g., natural language, DNA, or protein sequences. The experiments demonstrate the improvement of our algorithm compared to recent approaches.
Links
Please Cite
- D. Weese,
M. H. Schulz,
P. Perner,
“Efficient String Mining under Constraints via the Deferred Frequency Index”, 2008-07.
cite this publication
Contact
For questions, comments, or suggestions please contact:
David Weese david.weese@fu-berlin.de