A Framework for Recommending Resource

Allocation Based on Process Mining

MICHAEL ARIAS

E R I C R O JA S

J O R G E M U N O Z- G A M A

M A RC O S S E P Ú LV E DA

A Framework for Recommending Resource Allocation Based on Process Mining

2

Outline

Introduction

Approach◦ General approach◦ Resource allocation request◦ Resource process cube and allocation metrics◦ Implementation

Experimental evaluation Future work Conclusions

A Framework for Recommending Resource Allocation Based on Process Mining

3

Outline

Introduction

Approach◦ General approach◦ Resource allocation request◦ Resource process cube and allocation metrics◦ Implementation

Experimental evaluation Future work Conclusions

A Framework for Recommending Resource Allocation Based on Process Mining

4

Motivation

Resources

Business

processes

Cost reductio

n

Improve performa

nce

A Framework for Recommending Resource Allocation Based on Process Mining

5

Motivation

A Framework for Recommending Resource Allocation Based on Process Mining

6

Related work

Human resource allocation: an important issue in Business Process Management

Resource patterns

[1] Russell et al. 2005

Machine learning

[5] Liu et al. 2008Data mining

[3] Huang et al. 2011[4] Liu et al. 2012

[2] Rinderle-Ma and van der Aalst. 2007

Markov models

[6] Huang et al. 2011[7] Koschmider et al. 2011

Preference model

[8] Cabanillas et al. 2013

A Framework for Recommending Resource Allocation Based on Process Mining

7

Related work

[1] [2] [5] [6] [7] [3] [4] [8] Proposed

Activity profile ✓ ✓

Resource profile ✓ ✓ ✓ ✓ ✓

Performance & quality ✓

Resource meta-model ✓ ✓ ✓ ✓

History ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Process mining tool ✓ ✓ ✓

Allocation at sub-process level ✓

A Framework for Recommending Resource Allocation Based on Process Mining

8

Related work

[9] Zhao, W., Zhao, X.: Process mining from the organizational perspective, 2014

Dynamically allocation [9]

Research challenges

At real-time

Real data with organizational models

Multi-factor

Flexible and extensible methods

Use different criteria

Multi-level Allocation an activity, process or sub-process level

A Framework for Recommending Resource Allocation Based on Process Mining

9

Related work

[1] [2] [5] [6] [7] [3] [4] [8] Proposed

Activity profile ✓ ✓

Resource profile ✓ ✓ ✓ ✓ ✓

Performance & quality ✓

Resource meta-model ✓ ✓ ✓ ✓

History ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Process mining tool ✓ ✓ ✓

Allocation at sub-process level ✓

A Framework for Recommending Resource Allocation Based on Process Mining

10

General idea

Resource profile

Performance & Quality

History

Process mining tool

Activity profile

Allocation at sub-process level

Allocate the most appropriate resource

Framework for recommending resource allocation

Resource meta-model

A Framework for Recommending Resource Allocation Based on Process Mining

11

Outline

Introduction

Approach◦ General approach◦ Resource allocation request◦ Resource process cube and allocation metrics◦ Implementation

Experimental evaluation Future work Conclusions

A Framework for Recommending Resource Allocation Based on Process Mining

12

Proposed framework

Resource Allocation Request

BPA

Best Position Algorithm

Recommender System

Characteristics

Information

Weights

Event log

Human resources information

Resource Process Cube

1

A Framework for Recommending Resource Allocation Based on Process Mining

13

Proposed framework

Resource Allocation Request

BPA

Best Position Algorithm

Recommender System

Characteristics

Information

Weights

Event log

Human resources information

Resource Process Cube

2

A Framework for Recommending Resource Allocation Based on Process Mining

14

Proposed framework

Resource Allocation Request

BPA

Best Position Algorithm

Recommender System

Characteristics

Information

Weights

Event log

Human resources information

Resource Process Cube

3

A Framework for Recommending Resource Allocation Based on Process Mining

15

Proposed framework

Resource Allocation Request

BPA

Best Position Algorithm

Recommender System

Characteristics

Information

Weights

Event log

Human resources information

Resource Process Cube

4

A Framework for Recommending Resource Allocation Based on Process Mining

16

Proposed framework

Resource Allocation Request

BPA

Best Position Algorithm

Recommender System

Characteristics

Information

Weights

Event log

Human resources information

Resource Process Cube

5

A Framework for Recommending Resource Allocation Based on Process Mining

17

Proposed framework

Resource Allocation Request

BPA

Best Position Algorithm

Recommender System

Characteristics

Information

Weights

Event log

Human resources information

Resource Process Cube

1

2

3

4 5

A Framework for Recommending Resource Allocation Based on Process Mining

18

Help-Desk process

Help-Desk process

Printer Support

Server Support

Database

Support

Desktop Support

Help Desk

First-contact level

Ring…!!!I have a issue

A Framework for Recommending Resource Allocation Based on Process Mining

19

Resource allocation request

Resource Allocation Request

Characteristics

Information

Weights

1

A Framework for Recommending Resource Allocation Based on Process Mining

20

Resource allocation request

Characterize the request

Characterization c = (f1, …, fn)

c = sub-process, typology

Characteristics In the Help-Desk:

First contact level 1

Expert level 2

Printer-support

Server-supportTypologies

A Framework for Recommending Resource Allocation Based on Process Mining

21

Resource allocation request

Resource Allocation Request

Characteristics

Information

Weights

Event log

Human resources information2

Resource Process Cube

3

1

A Framework for Recommending Resource Allocation Based on Process Mining

22

Resource allocation request

Resource Allocation Request

Characteristics

Information

Weights

Event log

Human resources information2

Resource Process Cube

3

1

A Framework for Recommending Resource Allocation Based on Process Mining

23

Resource allocation request

Characterize the request

Proposed dimensions

Expertise matrices

Information

Frequency Performance QualityCostWorkload

Expertise Dimension

Competencies

Dimensions (history)

Resource process cube

(Q)

Skills

Knowledge

Ec [1 : n]

Er [1 : n]

D

Example

E c1 = [2,2]Er1 = [0,1]

Required Expertise

Resource Expertise

C1 = (sub-process 1,

printers)Required: Resource 1:

A Framework for Recommending Resource Allocation Based on Process Mining

24

Resource process cube (Q)

R1R2

Rn

…

Characte

rist

ics

Resource

s

Dimensions

Frequency

Performance

Quality

Cost

Workload

Sub-process,

typology

A Framework for Recommending Resource Allocation Based on Process Mining

25

Resource allocation metrics

Performance_Metric (r,c) =

[ ][c][ p ].max - [r][c][ p ].avg Q Q [ ][c][ p ].max - [ ][c][ p ].min

QQ

Prioritize the fastest resources

A Framework for Recommending Resource Allocation Based on Process Mining

26

Resource allocation metrics

Quality_Metric (r,c) =

[r][c][ q ].avg - [ ][c][ q ].min Q Q [ ][c][ q ].max - [ ][c][ q ].min

QQ

Prioritize the best evaluated resources

A Framework for Recommending Resource Allocation Based on Process Mining

27

Resource allocation metrics

Cost_Metric (r,c) =

[ ][c][ co ].max - [r][c][ co ].avg Q Q [ ][c][ co ].max - [ ][c][ co ].min

QQ

Prioritize the cheapest resources

A Framework for Recommending Resource Allocation Based on Process Mining

28

Resource allocation metrics

Workload_Metric (r,c) =

[r][ ][ w ].top - [r][ ][ w ].total Q Q [r][ ][ w ].top - [r][ ][ w ].bottom

QQ

Prioritize the most idle resources

A Framework for Recommending Resource Allocation Based on Process Mining

29

Resource allocation metrics

Frequency_Metric (r,c) =

logarithm ( [r][c][

f ].total) +1Q

logarithm ( [ ][c][ f ].total) +1

Q

Prioritize the most experienced resources

A Framework for Recommending Resource Allocation Based on Process Mining

30

Resource allocation metrics

1

- 1n√∑ n

(under(i))2

i =1

underQualification_Metric =

under(i) =

c[i]-

r[i] c[i] -

T

iE E

Eif c[i] >= r[i]E E

0 otherwise

Identify how below the resource is from the desired expertise level

A Framework for Recommending Resource Allocation Based on Process Mining

31

Resource allocation metrics

1

- 1n√∑ n

(over(i))2

i =1

overQualification_Metric =

over(i) =

r[i]-

c[i] -

c[i]

TiE E

Eif r[i] >= c[i]E E

0 otherwise

Identify how above the resource is from the desired expertise level

A Framework for Recommending Resource Allocation Based on Process Mining

32

Resource allocation request

Resource Allocation Request

Characteristics

Information

Weights

Event log

Human resources information2

Resource Process Cube

3

1

A Framework for Recommending Resource Allocation Based on Process Mining

33

Resource allocation request

Characterize the request

Describe the importance of each dimension

WeightsLarge size company:

F:010

P:050

Q:010 C:100

U:015 O:000

Small size company:

F:025

P:015

Q:100 C:030

U:075 O:065

A Framework for Recommending Resource Allocation Based on Process Mining

34

Proposed framework

Resource Allocation Request

BPA

Best Position Algorithm

Recommender System

Characteristics

Information

Weights

Event log

Human resources information2

Resource Process Cube

3

1

4 5

A Framework for Recommending Resource Allocation Based on Process Mining

35

Implementation with BPA [10]

Frequency

Data Item Local Score

R6 0.601

R4 0.593

R5 0.554

R1 0.525

… …

Performance

Data Item Local Score

R3 0.851

R2 0.833

R1 0.832

R5 0.775

… …

Quality

Data Item Local Score

R9 0.913

R7 0.864

R8 0.808

R2 0.723

… …

…

… …

… …

Position

1

2

3

4

…

…

… …

… …

…

… …

… …

Overall topData item Score

R1 0.795R2 0.788

R14 0.784

[10] Akbarinia, R., Pacitti, E., Valduriez, P.: Best position algorithms for ecient top-k query processing, 2011

A Framework for Recommending Resource Allocation Based on Process Mining

36

Outline

Introduction

Approach◦ General approach◦ Resource allocation request◦ Resource process cube and allocation metrics ◦ Implementation

Experimental evaluation Future work Conclusions

A Framework for Recommending Resource Allocation Based on Process Mining

37

Experimental evaluation

Office 365 Enterprise • Range of offers by user

type

• Advanced admin tools

• Enterprise IT capabilities

Enterprise

Help-Desk process

Preliminary evaluation

Case attributes Case ID, Sub-process group, Process typologyResource, Cost, Customer satisfaction (quality)Creation date, Closing date, Priority

Experiment 1Calculate the top 3-queries processing over the sorted lists, considering each single metric by itself.

Experiment 2

Experiment 3

3 types of companies: large size, mid size, and small size

Different event log sizes and the amount of resources were increased

38A Framework for Recommending Resource Allocation Based on Process Mining

Top 3-queriesSingle metric

Experiment 1

Experiments results

39A Framework for Recommending Resource Allocation Based on Process Mining

Top 3-queries3 help desk companies

Experiment 2

Experiments results

40A Framework for Recommending Resource Allocation Based on Process Mining

Top 3-queries3 help desk companiesLog sizeAmount of resources

Experiment 3

Experiments results

41A Framework for Recommending Resource Allocation Based on Process Mining

Performance analysis

( a ) Number of cases (thousands)

( b ) Number of resources

Experiments results

A Framework for Recommending Resource Allocation Based on Process Mining

42

Outline

Introduction

Approach◦ General approach◦ Resource allocation request◦ Resource process cube and allocation metrics◦ Implementation

Experimental evaluation Future work Conclusions

A Framework for Recommending Resource Allocation Based on Process Mining

43

Future work

Case studies using real data

Incorporate new dimensions

Explore potential application domains

Combine our approach with existing works

A Framework for Recommending Resource Allocation Based on Process Mining

44

Outline

Introduction

Approach◦ General approach◦ Resource allocation request◦ Resource process cube and allocation metrics◦ Implementation

Experimental evaluation Future work Conclusions

A Framework for Recommending Resource Allocation Based on Process Mining

45

Conclusions

Consider different perspectives

New allocation technique

Multi – factor criteria

Sub-process level

Fine-grained, generic, & extensible

Experimental evaluation

Synthetic dataBPA allows obtain a final ranking

THANK YOU FOR YOUR ATTENTION

A Framework for Recommending Resource

Allocation Based on Process Mining

MICHAEL ARIAS

E R I C R O JA S

J O R G E M U N O Z- G A M A

M A RC O S S E P Ú LV E DA

ANNEX 1

NEW DIMENSIONS

A Framework for Recommending Resource Allocation Based on Process Mining

49

New dimensions

ITAT by priority (Performance)

First Time Fix (Quality)

# of incidents by priority

(Compliance)

Resolved within SLA by priority

(Value)

% Reduction incidents

(Value)# of incidents resolved by…

(Quality)# of proactive problems

(Compliance)Average time to provide workaround by priority

(Efficiency)

Incident management

dimension

Problem management

dimension

ANNEX 2

BEST POSITION ALGORITHM

A Framework for Recommending Resource Allocation Based on Process Mining

51

Best Position Algorithm (BPA)

A Framework for Recommending Resource Allocation Based on Process Mining

52

BPA (iteration 1)

- Sorted Access to Pos 1:

{d1, d2, d3}

- Random Access to obtain local score y position.

P11= {1, 4, 9} and Bp1 = 1

P12= {1, 6, 8} and Bp2 = 1

P13= {1, 5, 8} y and Bp3 = 1

Best position score:

λ = f (S1 (1) + S2 (1) + S3 (1))

= 30 + 28 + 30 = 88

A Framework for Recommending Resource Allocation Based on Process Mining

53

BPA (iteration 1)

L1 P1 = {1, 4, 9} and BP1 = 1

L2 P2 = {1, 6, 8} and BP2 = 1

L3 P3 = {1, 5, 8} and BP3 = 1

λ = f (S1(1), S2(1), S3(1)) = 30 + 28 +30 = 88. λ be the ThresholdY = {d1, d2, d3}, Y visited elements Z = {65, 63, 70}, Z score global

Best position score = 88 > overall score of the highest data items {d1, d2, d3} (65,63,70) → continue

A Framework for Recommending Resource Allocation Based on Process Mining

54

BPA (iteration 2)

- Sorted Access to Pos 2:

{d4, d5, d6}

- Random Access to obtain local score and position.

P21= {2,7,8} and Bp21 = 2

P22= {2,4,9} and Bp22 = 2

P23= {2,4,6} and Bp23 = 2

Best position score:

λ = f (S1 (2) + S2 (2) + S3 (2))

= 28 + 27 + 29 = 84

A Framework for Recommending Resource Allocation Based on Process Mining

55

BPA (iteration 3)

- Sorted Access to Pos 3:

{d9, d7, d8}

- Random Access to obtain local score and position.

P31= {3,5,6} and Bp31 = 9

P32= {3,5,7} and Bp32 = 9

P33= {3,9,10} and Bp33 = 6

Best position score:

λ = f (S1 (9) + S2 (9) + S3 (6))

= 11 + 13 + 19 = 43

A Framework for Recommending Resource Allocation Based on Process Mining

56

BPA (iteration 3)P31 = {3, 5, 6} Bp1= 9, P32 = {3, 5, 7} Bp2= 9, P33 = {3, 9, 10} Bp3= 6

PL1 = PIt1L1U PIt2L1 U PIt3L1= {1,2,3,4,5,6,7,8,9}

PL2 = PIt1L2U PIt2L2 U PIt3L2= {1,2,3,4,5,6,7,8,9}

PL3 = PIt1L3U PIt2L3 U PIt3L3= {1,2,3,4,5,6,8,9,10}

λ = f (S1(9), S2(9), S3(6)) = 11 + 13 + 19 = 43

Last Global Score: {d3, d4, d6} = {70, 66, 70}

Current Global Score : {d9, d7, d8} = {62, 61, 71}

Update Y. Y = {d3, d6, d8}, = {70, 70, 71}

The values in Y are greater than λ= 43, the stop condition finish in position 3.

A Framework for Recommending Resource Allocation Based on Process Mining

57

BPA (iteration 3)