Information Technology Reference
In-Depth Information
5 Related Work
Many literatures have targeted in the similarity search of process models. Some
propose creative algorithms for measuring the similarity between business pro-
cess models [12][13][14][15][16]. Some aim at similarity comparison on state ma-
chines [17][18]. In [12] a combination of structural properties and label similarity
is adopted to compare process models. However it doesn't define an effective
fuzzy query metric which is meaningful in the real-life applications. Li [13] com-
pares process models based on the model transformation technique. See [16] a
graph edit distance is utilized as the metric for process model comparison. Some
focus on the process equivalence and have different judgments. In context of pro-
cess mining [19], [20] proposes a use of ”typical behavior” recorded in event logs.
Similar work like equivalence notions, trace equivalence [21] and bi-simulation
[22] do not take the syntactical structure into consideration. Moreover, these ap-
proaches actually not consider the degree of similarity which is more important
in practice. In work of [23], the researchers evaluated several similarity met-
rics and concluded that a structural similarity metric based on graph matching
brings highest retrieval quality. In their follow-up paper [24], four graph match-
ing algorithms are presented. In these algorithms only 1-to-1 correspondences
between nodes in the compared process models are established. An improved
investigation is shown in [25] that calculate 1-to-N or N-to-M correspondences.
Still, all these graph matching algorithms need to strike a tradeoff between the
retrieval quality and the time complexity.
This paper is the first to propose a causal similarity search approach by first
calculating correspondence sets and pruning on them on the basis of structural
properties. In addition, we introduce a fuzzy similarity search metric on model
patterns that is more reasonable in practical applications. Moreover, we replace
the complex graph traversal process with a series of look-up steps in adjacency
matrixes, which reduces the complexity. All these three points make the work
different from the references mentioned in this section. What's more, the exper-
imental results show the validity of our approach for mining similar patterns
from pair of process models.
6Conluon
This paper presented a novel approach to find similar patterns from pairs of
process models: a combination of semantic comparison in labels and topological
consideration with adjacency matrix for the similarity search on general process
models. Also, it introduces a metric of fuzzy similarity search on model pat-
terns which is more reasonable in practical applications. In addition, we convert
the traditional graph traversal process to a series look-up steps in adjacency
matrixes, which reduces the complexity.
The approach has been tested on a small collection of process models and
proved the proposed approach is valid to some extent. However, there is much
work to do in the future. For example, the approach still needs further test in
