discovering social networks from event logs

21
Discovering Social Networks from Event Logs Data Mining TIE 522 KIM, HYUNKI Wil M.P van der Aalst 1 , Hajo A. Reijers 1 , Minseok Song 2,1

Upload: kyla-stanley

Post on 02-Jan-2016

46 views

Category:

Documents


2 download

DESCRIPTION

Discovering Social Networks from Event Logs. Wil M.P van der Aalst 1 , Hajo A. Reijers 1 , Minseok Song 2,1. Data Mining TIE 522 KIM, HYUNKI. Contents. ■ Introduction. ■ The concept of process mining. ■ Mining organizational relations. ■ Metrics. ■ MiSoN. ■ Case study. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST

Discovering Social Networks from Event Logs

Data MiningTIE 522

KIM, HYUNKI

Wil M.P van der Aalst1, Hajo A. Reijers1, Minseok Song2,1

Page 2: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST

Contents

■ Introduction

■ MiSoN

■ The concept of process mining

■ Mining organizational relations

■ Metrics

■ Case study

2

■ Conclusion

Page 3: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST3

Introduction

▣ Two concepts

• Business Process Management(i.e., systems supporting the design, analysis, and enactment of opera-

tional business processes)

• Sociometry: refered to as sociography, refers to methods presenting data on interpersonal relation-

ships in graph or matrix form.

▣ SNA(Social Network Analysis)

: refers to the collection of method, techniques and tools in sociometry aiming at the analysis of social

networks.

▣ This paper presents the approach, the various metrics that can be used to build a social network, Mi-

SoN(Mining Social Networks), and case study.

▣ The main challenge is to derive social networks from this type of data such as today’s enterprise in-

formation systems.

Page 4: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST4

Process mining

▣ The goal of process mining is to extract information about processes from transaction logs.

- An event log -

▣ Sociograms can be used as input for SNA tools that can visualize the network in various

way.

(compute metrics like the density of the network, analyze the role of an individual in the network, and

identify cliques)

Discovering social networks

- The sociogram based on the event log -

Page 5: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST5

Process mining

▣ α-algorithm

- A process model based onα-algorithm -

▣ The presence of timing information and information on cases/activities allows for more advanced

forms of process mining.

Other types of mining

▣ Another interesting application of process mining is fraud detection.

Page 6: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST6

Mining organizational relations

▣ Motivations

1. Existing systems record information about human activity

2. There is a wide variety of mature techniques and tools to analyze such sociograms.

Social network analysis

▣ Two approaches

1. Sociocentric approach

: consider interactions within a defined group and consider the group as a whole

2. Egocentric approach

: consider the network of an individual(relations among the friends of a given person).

▣ Graph notation

• can be undirected or directed.

• the relations may be binary or weighted. The weight is used to qualify the relation.

Page 7: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST7

Mining organizational relations

▣ How to derive meaningful sociograms from event log?

Deriving relations from event logs

▣ Four types of metrics

1. Metrics based on (possible) causality

: monitor for individual cases how work moves among performers.(handover of work)

2. Metris based on joint cases

: ignore causal dependencies but simply count how frequently two individuals are performing activities

for the same case.

3. Metrics based on joint activities

: do not consider how individuals work together on shared cases but focus on the activities they do.

4. Metrics based on special event types

: consider the type of event.

Page 8: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST8

Metrics

▣ Convenient notation for event logs

• Definition 4.1. (Event log)

A: a set of activities

P: a set of performers

E: a set of possible event(E = A x P)

C: the set of possible event sequences(C = E)

L∋В(C) is the set of all bags(multi-sets) over C

• First, abstract from additional information such as time stamps, data, etc.

• Second, do not consider the ordering of events corresponding to different cases.

• For convenience, we define two operations on events: πa(e) = a and πp(e) = p for some event e =

(a,p).

Page 9: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST9

MetricsMetrics based on (possible) causality

▣ Metrics based on causality take into account both handover of work and sub-contracting.

▣ Three kinds of refinements

1. One can differentiate with respect to the degree of causality.(length of handover)

2. We can ignore multiple transfers within one instance or not

3. We can consider arbitrary transfers of work or only consider those where there is a casual depen-

dency

- A process model based onα-algorithm -

:the function which returns true if within the context of case c performers p1 and p2 both executed

some activity such that the distance between these two activities is n.:the function which returns the number of times in the case c.

is the same as ?

Page 10: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST10

MetricsMetrics based on (possible) causality

Page 11: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST11

MetricsMetrics based on (possible) causality

Page 12: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST12

MetricsMetrics based on (possible) causality

Page 13: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST13

MetricsMetrics based on joint cases

MetricsMetrics based on joint activities

▣ Measurement of the distance be-

tween two performers.

• Minkowski distance Hamming dis-

tance

• Pearson’s correlation coefficient

Page 14: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST14

MetricsMetrics based on special event types

▣ This method considers event types such as schedule, assign, withdraw, reassign, start,

suspend, resume, pi-abort, ate-abort, complete, autoskip, manualskip, and unknown.

: the function which returns true if within the context of the case c performers p1 and p2 both

executed the same activity and p1 was responsible for a specific type of event and p2 is

the first performer of some event for the same activity: the function which returns the number of times in the case c.

Page 15: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST15

MiSoN(Mining Social Networks)

▣ MiSoN has been developed to discover relationships between individuals from a range of

enterprise information systems including workflow management systems.

▣ MiSoN constructs sociograms that can be used as a starting point for SNA.

Page 16: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST16

MiSoN(Mining Social Networks)

▣ MiSoN

• provides functionalities for displaying user statistics and event log statis-

tics.

• the user can select suitable metrics and set relevant options.

• can export the mining result using the AGNA-translator.(AGNA: a SNA tool)

• can export the mining result to other SNA tools like UCINET and NetMiner.

Page 17: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST17

Case study

Type of event

Matrix view

- MiSoN screenshot of ProM-

Page 18: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST18

Case study

▣ The result from the handover of work.

Page 19: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST19

Case study

▣ The result from the working together – considering distance with causality.

Page 20: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST20

Case study

▣ The result from the similar task– considering Hamming distance.

Page 21: Discovering Social Networks from Event Logs

CSI Lab. DHE. UNIST21

Conclusion

▣ This paper is to present an approach to discover sociograms.

▣ MiSoN can interface with commercial systems such as Staffware and standard SNA tools

like AGNA and UCINET and NetMiner.

▣ Two kinds of limitations.

1. Only mornitoring the events.

2. Enforcing the interaction patterns.

▣ Future research

1. We plan to repeat our analysis within the public works department and apply our approach in many

other organizations as well.

2. we also investigate extensions of the approach using filtering techniques and more advanced forms

clustering.