The assessment of behavioral rules with respect to a given dataset is key in several research areas, including declarative process mining, association rule mining, and specification mining. The assessment is required to check how well a set of discovered rules describes the input data, as well as to determine to what extent data complies with predefined rules. In declarative process mining, in particular, some measures have been taken from association rule mining and adapted to support the assessment of temporal rules on event logs. Among them, support and confidence are used more often, yet they are reportedly unable to provide a sufficiently rich feedback to users and often cause spurious rules to be discovered from logs. In addition, these measures are designed to work on a predefined set of rules, thus lacking generality and extensibility. In this paper, we address this research gap by developing a general measurement framework for temporal rules based on Linear-time Temporal Logic with Past on Finite Traces (LTLpf). The framework is independent from the rule-specification language of choice and allows users to define new measures. We show that our framework can seamlessly adapt well-known measures of the association rule mining field to declarative process mining. Also, we test our software prototype implementing the framework on synthetic and real-world data, and investigate the properties characterizing those measures in the context of process analysis.
A Temporal Logic-Based Measurement Framework for Process Mining / Cecconi, Alessio; De Giacomo, Giuseppe; Di Ciccio, Claudio; Maggi, Fabrizio Maria; Mendling, Jan. - (2020), pp. 113-120. (Intervento presentato al convegno 2nd International Conference on Process Mining, ICPM 2020 tenutosi a Padua, Italy) [10.1109/ICPM49681.2020.00026].
A Temporal Logic-Based Measurement Framework for Process Mining
De Giacomo, Giuseppe
;Di Ciccio, Claudio
;Maggi, Fabrizio Maria
;
2020
Abstract
The assessment of behavioral rules with respect to a given dataset is key in several research areas, including declarative process mining, association rule mining, and specification mining. The assessment is required to check how well a set of discovered rules describes the input data, as well as to determine to what extent data complies with predefined rules. In declarative process mining, in particular, some measures have been taken from association rule mining and adapted to support the assessment of temporal rules on event logs. Among them, support and confidence are used more often, yet they are reportedly unable to provide a sufficiently rich feedback to users and often cause spurious rules to be discovered from logs. In addition, these measures are designed to work on a predefined set of rules, thus lacking generality and extensibility. In this paper, we address this research gap by developing a general measurement framework for temporal rules based on Linear-time Temporal Logic with Past on Finite Traces (LTLpf). The framework is independent from the rule-specification language of choice and allows users to define new measures. We show that our framework can seamlessly adapt well-known measures of the association rule mining field to declarative process mining. Also, we test our software prototype implementing the framework on synthetic and real-world data, and investigate the properties characterizing those measures in the context of process analysis.File | Dimensione | Formato | |
---|---|---|---|
Cecconi_A-Temporal_2020.pdf
solo gestori archivio
Tipologia:
Versione editoriale (versione pubblicata con il layout dell'editore)
Licenza:
Tutti i diritti riservati (All rights reserved)
Dimensione
306.57 kB
Formato
Adobe PDF
|
306.57 kB | Adobe PDF | Contatta l'autore |
Cecconi_postprint_A-Temporal_2020.pdf
accesso aperto
Note: DOI: 10.1109/ICPM49681.2020.00026
Tipologia:
Documento in Post-print (versione successiva alla peer review e accettata per la pubblicazione)
Licenza:
Tutti i diritti riservati (All rights reserved)
Dimensione
547.17 kB
Formato
Adobe PDF
|
547.17 kB | Adobe PDF |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.