Adding temporal information has been the focus of research for years in the pattern mining field since it is highly relevant to accurately describe medical treatments, natural disasters or customer shopping sequences, among others. The sequential pattern mining task [Mab10] has been widely used to cope with problems in which events or elements are ordered over time. However, this task lacks in providing information on the right time events occur, and it just describes whether an event takes play before/after another. The specialized literature recently includes research works in which the time variable determines any periodicity among sequences [Fou19]. This analysis enables to find out cyclic sequences, discarding those having a time elapsed that is too large.
In late 2017, temporal pattern mining underwent important improvements with the proposal of temporal constraint networks, also known as chronicles [Dau17]. A chronicle represents a set of events linked by quantitative temporal constraints, and it is essential to describe, for example, the time elapsed among maintenance events to avoid a system failure. However, chronicle mining is at a very incipient stage with just a couple of articles in the Computer Science literature and mainly focused on predictive purposes [Dau17] [Sel18]. Here, it is essential to redefine the chronicle mining concept from a descriptive point of view and to work on chronicles as paths fully satisfied in a sequence and not as unconnected pairs of linked events as it usually does.
Additionally, it is key to extract more user-centric knowledge so it can be easily applicable and focused on the task at hand (increasing the actionability).
The problem of mining chronicles is relatively new, and it has not been deeply studied and can be improved in the following ways:
- Chronicle mining as a descriptive task in which useful temporal constraint networks are discovered according to some user’s constraints, e. g. min/max elapsed time, a minimum number of events, among others. Here, it is also important to define a chronicle as a way of describing fully satisfied sequences of events and not just pairs of linked events as it usually does.
- Development of algorithms for mining chronicles including the following features: capacity to reach optimal solutions with small computational and memory consumptions; ability to constraint the solutions according to the users’ aim; feasibility to introduce subjective knowledge based on the users’ experience.
- Design of chronicle mining algorithms that considers the utility of the events so not only the temporal constraints are born in mind, but the final utility (utility maximization) achieved from the mined chronicles. Periodicity among sequences is also key as a way of discarding isolated or acyclic sequences that should not be considered.
The research group has an important background in the pattern mining field [Ven16] and, recently, has developed some proposals to extract local periodic patterns [Fou21] as well as user-centric knowledge [Lun20].
[Mab10] N. R. Mabroukeh, C. I. Ezeife: A taxonomy of sequential pattern mining algorithms. ACM Comput. Surv. 43(1): 3:1-3:41 (2010).
[Fou19] P. Fournier-Viger, Z. Li, J. Chun-Wei Lin, R. U. Kiran, H. Fujita: Efficient algorithms to identify periodic patterns in multiple sequences. Inf. Sci. 489: 205-226 (2019).
[Dau17] Y. Dauxais, T. Guyet, D. Gross-Amblard, A. Happe: Discriminant Chronicles Mining – Application to Care Pathways Analytics. AIME 2017: 234-244
[Sel18] C. Sellami, A. Samet, M. A. B. Tobji: Frequent Chronicle Mining: Application on Predictive Maintenance. ICMLA 2018: 1388-1393.
[Ven16] S. Ventura, J. M. Luna: Pattern Mining with Evolutionary Algorithms. Springer 2016, ISBN 978-3-319-33857-6, pp. 1-190.
[Fou21] P. Fournier-Viger, P. Yang, R. U, Kiran, S. Ventura, José M. Luna: Mining local periodic patterns in a discrete sequence. Information Science. 544: 519-548 (2021).
[Lun20] J. M. Luna, P. Fournier-Viger, S. Ventura: Extracting User-Centric Knowledge on Two Different Spaces: Concepts and Records. IEEE Access 8: 134782-134799 (2020).