SmartFlow: Mining and Optimizing Ad Hoc Workflows

Ad hoc workflows are everywhere in service industry, scientific research, as well as daily life, such as the workflow of customer service, problem solving, information searching, expert finding, and decision making. Optimizing ad hoc workflows thus has significant benefits to the society. Currently the execution of ad hoc workflows is based on human decisions, where misinterpretation, inexperience, and ineffective processing are not uncommon, leading to operation inefficiency.

 

The goal of this project is to design and develop fundamental models, concepts, and algorithms to mine and optimize ad hoc workflows. An ad hoc workflow typically consists of impromptu processes that are determined dynamically by individual agents, based on the nature of the workflow, the expertise of the agents, as well as the interaction among the agents. There are three research challenges that have not been addressed systematically in existing literature. First, what are the appropriate models to represent and characterize ad hoc workflows? Second, given these models, how to mine and optimize ad hoc workflows? Third, what are the social and business implications of these mining results? In this project we will address these challenges and provide a comprehensive study of ad hoc workflow mining and optimization. Specifically, three technical themes are identified. (1) Network Modeling and Structure Mining. A network model is built that statistically captures the execution characteristics of ad hoc workflows, and is optimized to improve the execution of new workflows with respect to different optimization objectives. (2) Workflow Artifact Mining. The network model built on workflow executions is then extended with workflow artifact mining to realize an optimization system that is able to take advantage of both executions and text contents. (3) Role Discovery and Relation Assessment. A computational framework is built to quantitatively analyze the roles and relationships of agents involved in ad hoc workflow executions in order to further optimize workflows.

Advances from this project will include models to represent ad hoc workflows, algorithms for mining hidden collaborative models, and techniques that optimize ad hoc workflow processing. The project bridges two emerging research areas, service science and network science, and enriches the principles and technologies of data mining.

 

 

Faculty:

            Yi Chen < yi.chen at njit dot edu >

Students:

             Peng Sun

             Ziyang Liu

             Yi Shan

             Brian Ackerman

Collaborator:

             Xifeng Yan (UCSB)(with an NSF collaborative award IIS-0917228),

             Shu Tao (IBM)

 

This project is supported by NSF IIS-0915438 and an IBM faculty award.