Information Technology Reference
In-Depth Information
Q-learning Algorithm for Task Allocation
Based on Social Relation
Xingmei Liu, Jian Chen, Yu Ji, and Yang Yu
School of Information Science and Technology
Sun Yat-Sen University
Guangzhou, Guangdong, P.R.China
Abstract.
Social relation has a great impact on task allocation of a
workflow management system. Most methods only focus on analyzing the
effect of the handover relation. However, a workflow is a team process
since many resources work on various tasks together to complete an
instance. So the influence of the previous resources (called SR) should
be considered during dynamic task allocation. This paper proposes a
method to compute the social relation between two resources. Then we
present a model to capture the influence of the previous resources on the
candidate resources and implement a Q-learning algorithm for dynamic
task allocation based on flow time perspective. Comparison experiments
show that the algorithm with SR has almost 53% improvement of a
real data set and 40% improvement of a simulation data set in flow time
perspective. There is a higher throughput than the algorithm without SR.
Experiment results confirm the existence of such relationship and prove
that it is necessary to consider the influence of all previous resources
during dynamic task allocation.
Keywords:
task allocation, social relation, markov decision process,
Q-learning.
1
Introduction
Task allocation is an important issue in Business Process Management. The
workflow engine should allocate tasks to appropriate resources in order to im-
prove the performance of business process. Resources can be divided into human
resources and non-human resources in workflow management system [1]. And
there are many factors influencing the human resources behavior. For example,
the workload of the human will affect the humans ability, namely Yerkes-Dodson
Law of Arousal, which shows that a worker will take less time to execute an task
if he/she is under some work pressure [2]. However, if the pressure is too high,
the worker's performance may degrade. And the social relation between two re-
sources also has influence on work eciency. However, most papers only focus
on the handover relation which means the candidate resources who will perform
successive task will be affected by the resource who performed the precursor task
[3]. In reality, many people are involved in a task at the same time, just like a
