Using Graph Aggregation for Service Interaction Message Correlation - Advanced Information Systems Engineering - page 640

Using Graph Aggregation for Service Interaction
Message Correlation

Adnene Guabtni 1 , 2 , Hamid Reza Motahari-Nezhad 3 , and Boualem Benatallah 1

1 The University of New South Wales, Sydney, Australia
{ aguabtni, boualem } @cse.unsw.edu.au
2 National ICT Australia (NICTA), Sydney, Australia
[e-mail]
3 HP Labs, Palo Alto, CA, USA
[e-mail]

Abstract. Discovering the behavior of services and their interactions in

an enterprise requires the ability to correlate service interaction messages
into process instances. The service interaction logic (or process model)
is then discovered from the set of process instances that are the result
of a given way of correlating messages. However, sometimes, the Cor-
relation Conditions (CC) allowing to identify correlations of messages
from a service interaction log are not known. In such cases, and with
a large number of message's correlator attributes, we are facing a large
space of possible ways messages may be correlated which makes identi-
fying process instances dicult. In this paper, we propose an approach
based on message indexation and aggregation to generate a size-ecient
Aggregated Correlation Graph (ACG) that exhibits all the ways mes-
sages correlate in a service interaction log not only for disparate pairs
of messages but also for sequences of messages corresponding to process
instances. Adapted filtering techniques based on user defined heuristics
are then applied on such a graph to help the analysts eciently identify
the most frequently executed processes from their sequences of CCs. The
approach has been implemented and experiments show its effectiveness
to identify relevant sequences of CCs from large service interaction logs.

Keywords: SOA, Process mining, Correlation, Aggregation.

1 Introduction

As the number of services in organizations are growing and service interactions
are getting more dynamic, there is a significant interest in understanding the re-
lationships and behavior of services in the enterprise. Approaches for discovering
the behavior of systems and services (also known as process or workflow discovery
[10, 4]) take process instances as input. A process instance is a sequence of service
messages corresponding to one execution of a process model. In the context of
services, a process instance consists of a sequence of messages that are exchanged
by services. However, identifying process instances, i.e. correlating messages so
that we know which service messages belong to the same process instance, is not