Transcript Slide 1

Essays on using Formal Concept Analysis in
Information Engineering
AA
Jonas Poelmans, Aspirant FWO
Katholieke Universiteit Leuven
Inno.com February 2011
Outline
1. Formal Concept Analysis
2.
FCA in data mining
2.1 Literature study
2.2 Domestic violence case study
3. FCA for mining temporal data
3.1 Human trafficking
3.2 Terrorist threat assessment
3.3 Integrated care pathways
4. Future research
Formal Concept Analysis (FCA)
•
•
•
•
Initially a mathematical technique
Visually represents the concepts available in the data
Now: used as exploratory data analysis technique
Starting point: cross table
Rows: contain the objects
Columns: contain the attributes
Crosses: relationships between the objects and the attributes
FCA essentials: cross table
Report 1
kicking
dad hits
me
X
X
Report 2
Report 3
X
X
stabbing
cursing scratching maltreating
X
X
X
X
X
X
X
Report 4
Report 5
X
X
X
FCA essentials: concepts
• Distill concepts from this cross table
• Concept: 2 parts
extent: reports that belong to the concept
intent: terms in reports
• Example:
attributes of report 5: “cursing”, “scratching” (= set A)
collect reports containing these terms: reports 2, 3, 5 (= set O)
=> Concept = (O,A)
FCA essentials: ordering of concepts
• Subconcept-superconcept relation:
concept d is subconcept of concept e if extent of d is
subset of extent of e
• Example:
Concept A with intent “cursing”, “scratching”, “stabbing” is
subconcept of concept B with intent “cursing”, “scratching”
A = {report 2, report 3}
B = {report 2, report 3, report 5}
FCA essentials: lattice
Literature study
• 702 papers on FCA published between 2003-2009
• FCA is used to:
cluster papers
visually represent clusters and their relationships
explore data
• Cross table:
objects = papers
attributes = terms
crosses = terms occuring in abstract of papers
Literature study: FCA papers
Literature study: KDD papers
Domestic violence case study
•
•
•
•
Many theoretical papers
Relatively few real life applications
Now: FCA used as exploratory data analysis technique
Aim: exploring and refining the concept of domestic violence
to improve detection & handling of domestic violence cases
Emergent Self Organising Map (ESOM)
ESOM:
–
–
–
–
Topographic map
Visualizing sparse high-dimensional datasets
Now: used as exploratory data analysis technique
Aim: exploring and refining the concept of domestic violence to
improve detection and handling of domestic violence cases
Concept - Knowledge theory (C-K theory)
Motivation
• 1997: first inquiry into nature and scope of domestic violence.
• 45% of population once fell victim to non-incidental domestic
violence.
• 27% of population: incidents occurred on a weekly or daily
basis.
• Pivotal project of Prime Minister Balkenende administration.
Dataset
Selection of 4814 police
reports from year 2007
Collection of terms: “kicking”,
“mother”, “child”, etc.
Each report is associated with
a subset of terms.
Title of incident ….
Reporting of the crime
Last night I was attacked by my
husband. I was watching television in
the living room when he suddenly
attacked me with a knife. I fell on the
floor. Then he tried to kick me in my
stomach. I tried to escape through the
back door while I was yelling for help.
I ran to the neighbours for help. They
called the emergency services.
Meanwhile my son ran away. My leg
was bleeding; my head was bouncing,
etc.
Domestic violence definition
3 components:
– Perpetrator is member of domestic sphere of victim:
partners, ex-partners, family members, relatives, family
friends
– Act of violence: physical assault, sexual harassment,
threatening
– Dependency relationship between perpetrator and
victim
Clustering terms for FCA
Terms grouped together in clusters based on domestic violence
definition:
“kicking”, “scratching”, “maltreating”, etc.
– ‘acts of violence’
“aunt”, “uncle”, “nephew”, etc.
– ‘relatives’
“father”, “mother”, “son”, etc.
– ‘family members’
etc.
Expanding CC and Transforming CK
Last night I was attacked by my husband. I was watching
television in the living room when he suddenly attacked me
with a knife. I fell on the floor. Then he tried to kick me in my
stomach. I tried to escape through the back door while I was
yelling for help. I ran to the neighbours for help. They called
the emergency services. Meanwhile my son ran away. My
leg was bleeding; my head was bouncing, etc.
Yesterday morning I was taking a bath. Suddenly my
daughter ran into the bathroom followed by her ex-boyfriend.
She screamed for help. He had a gun in his hand and he was
clearly under influence of beer or drugs. He yelled out that he
couldn’t live without her. He threatened to kill me and my
daughter if she wouldn’t come back to their house. The
neighbours who were alarmed by all the noise came to give
some help. Meanwhile another neighbour phoned the police.
I jumped out of my bath and tried to push him on the floor.
During this fight I got some serious injuries on my back etc.
This morning I wanted to go shopping. While I opened the
door of my car my ex-husband jumped out of the bushes with
a baseball bat in his hand. He hit me several times in my
stomach, etc.
Transforming KC: Initial ESOM map
Expanding CC and Transforming CK
Last night I was attacked by my husband. I was watching
television in the living room when he suddenly attacked me
with a knife. I fell on the floor. Then he tried to kick me in my
stomach. I tried to escape through the back door while I was
yelling for help. I ran to the neighbours for help. They called
the emergency services. Meanwhile my son ran away. My
leg was bleeding; my head was bouncing, etc.
Yesterday morning I was taking a bath. Suddenly my
daughter ran into the bathroom followed by her ex-boyfriend.
She screamed for help. He had a gun in his hand and he was
clearly under influence of beer or drugs. He yelled out that he
couldn’t live without her. He threatened to kill me and my
daughter if she wouldn’t come back to their house. The
neighbours who were alarmed by all the noise came to give
some help. Meanwhile another neighbour phoned the police.
I jumped out of my bath and tried to push him on the floor.
During this fight I got some serious injuries on my back etc.
This morning I wanted to go shopping. While I opened the
door of my car my ex-husband jumped out of the bushes with
a baseball bat in his hand. He hit me several times in my
stomach, etc.
Expanding KK: Newly discovered features
Pepper spray
Homosexual relationship, lesbian relationship
Sexual abuse, incest
Alternative spelling of some words (e.g. ex-boyfriend, exboyfriend, ex boyfriend)
Violence terms lacking in the thesaurus: abduction, choke, strangle, etc.
Weapons lacking in the thesaurus: belt, kitchen knife, baseball bat, etc.
Terms referring to persons: partner, fiancée, mistress, concubine, man next
door, etc.
Terms referring to relationships: romance, love affair, marriage problems,
divorce proceedings, etc.
Reception centers: woman’s refuge center, home for battered woman, etc.
Terms referring to an extra marital affair: I have an another man, lover, I am
unfaithful, etc.
Expanding CC: Identifying faulty case labelling
Expanding CC: Testing prior knowledge
Expanding CC: Omitted information
Expanding CC: Inconsistencies
Expanding CC: Niche cases
Expanding KK: Omitted information and faulty case
labellings
-
Burglary cases
No suspect in domestic violence cases
Third person makes statement to police
Violence between a caretaker and an inhabitant of
an institution
Expanding KK: FCA classification rules
FCA results: classification rules
- 22 domestic violence and 15 non-domestic
violence classification rules: label automatically
and correctly 75% of incoming cases
FCA results: correct filed report labels
Transforming KC: Improved ESOM map
Expanding KK: Confusing situations for police officers
- Lover boys
- Extramarital relationship
- Violence between a caretaker and an inhabitant of an
institution
- Violence between colleagues
- An ex-boyfriend attacks the new boyfriend
- Third person makes statement to the police for somebody else
ESOM risk classification map
ESOM risk map results: label cases
- Label remaining 78% of incoming cases based on area of map
FCA & ESOM results
- Automatically label 91% of incoming cases
correctly
- Improved police training
- Upgraded domestic violence definition
- Identification of niche cases
- Data quality improvement
Conclusions
• Combination of FCA & ESOM showcased as exploratory
data analysis technique, with success
• Refinement of the concept ‘domestic violence’
• Accurate case classification model
Human trafficking
• recruitment, transportation, harboring and receipt of people for
slavery, forced labor and servitude
• fastest growing criminal industry in the world
• global annual market of 42.5 billion
• 700 000 to 2 million women and girls are trafficked across
international borders every year
Human trafficking (2)
• majority of victims are trafficked in commercial sexual
exploitation
• threats, violence, coercion, deception, abuse of power make
victim consent to exploitation
• victims of human trafficking rarely make official statement to
the police
• human trafficking team is installed to proactively search police
database for signals of human trafficking
Police challenges
• databases contain large amount of observational reports
• 10% of information containing human trafficking indicators is
labeled as such by police officers
• documents are spread over multiple database systems
• limited browsing functionality is provided.
• currently: officers have to search all these databases and
manually inspect reports for indications
Investigation procedure
• collect sufficient evidence and indications against potential
suspect
• construct document based on section 273f of the code of
criminal law
• send request to Public Prosecutor to start in-depth investigation
Dataset
• 69 788 general reports from 2008
• observations made by police officers during motor vehicle
inspections, police patrols, etc.
Temporal Concept Analysis (TCA)
TCA:
–
Based on FCA
–
Addresses the problem of representing time
–
Is particularly suited as for visual representation of discrete
temporal phenomena
Analysis method
• FCA:
– identify early warning indicators in police reports
– detect and extract potential trafficking suspects
• TCA:
– profiling potential suspects and their evolution over time
– gaining insight in their social network
Human trafficking indicators
Static indicators
Indicators with time dimension
Indicators coming from social network
- nationality
- red light district: Wallen
- seen with known suspect
- violence
- red light district: Ruysdaelkade
- minors involved
- regularly visiting suspicious club
- restriction of personal freedom
- regularly dropping of girls at club
- id-problems
- expensive car
- carrying large amount of money
- woman in car
- forcing person to work in bad condition
- prostitute involved
- dependency relationship
- car trade
- injury observed
- woman not speaking
Example of a formal context and report
Expensive
cars
Prostitues
Id-papers
Vacation
Former
eastern
Europe
1
X
X
X
X
x
2
x
x
x
X
3
X
X
x
X
4
5
x
x
x
X
Report 1:
On the night of 23 of march 2008 we
stopped a car with a Bulgarian
license plate for routine motor
vehicle inspection. It was a
Mercedes GLK with licence plate BL
XXX. The car was driving around in
circles in a prostitution area. On the
backseat of the car we noticed two
well dressed young girls. We asked
for their identification papers but
they didn’t speak English nor Dutch.
The driver of the car was in
possession of their papers and told
us that they were on vacation in the
Netherlands for two weeks etc.
Detecting possible suspects
Data table of a conceptual time system
Time part
Time granule
Date
Event part
Expensive car
0
26-1-2008
1
21-2-2008
2
15-2-2008
3
13-3-2008
X
4
27-4-2008
X
5
1-6-2008
6
14-6-2008
7
18-6-2008
Prostitution area
Vacation
X
X
X
X
X
X
X
X
X
X
Profiling suspects with TCA
Network evolution analysis using TCA
Terrorist threat assessment with Formal Concept
Analysis
• Terrorism in the Netherlands:
– the brute murder of the filmmaker Theo van Gogh.
• Introduction of the term “European Jihad” by the secret
services, most important trend is
– the evolvement from exogenous foreign terrorist threat to
indigenous home-grown terrorism
• The law on terrorism in the Netherlands allows to proactively
search for possible jihadists.
The four phase model of radicalism
The phases explained
Incremental isolation
• In the preliminary phase the subject experiences a crisis of
confidence.
• In the social alienation phase a small minority of these young
Muslims cannot handle this situation.
• In the Jihadization phase the subjects are characterized by
strong radical Islamic convictions and the fact that they
condone violence.
• The Jihad/Extremism is a phase of total isolation. The
subjects’ entire lives are governed by their radical Islamic
beliefs.
Indicators and terms
Anti western
Orthodox religion
Change behavior
Monkeys; pigs
soennah
Suddenly wearing beard
Imperialists; Zionists
fatwa; fatwah
no respect colleague
Kufar; kufir
wahabist
islamitic mariage
Unbeliever; disbeliever
Only and one islam
tradional clothes
takfir
fundamentalist
Not shaking hands women
Dataset
• 166577 general reports
– Contain observations made by police officers
– Consequence of implementing Intelligence Led Policing in 2005
– Growing each year:
• 2006: 41900
• 2007: 54799
• 2008: 69788
– The unstructured text describing the observations is explored
• The observations are made within the communities of
Amsterdam, Amstelveen, Uithoorn, OuderAmstel and
Diemen
Example police report
Example of a formal context for a subject
Subject
A
Anti western
Orthodox religion
Change
behavior
X
X
X
x
x
B
C
X
D
E
x
x
x
x
Report 1:
On the night of 23 of march 2008 we
stopped a car with three men with
traditional clothes. One man, named
[A], refused to speak to us and
called us monkeys….
Report 2:
On the evening of 6 of april 2008 we
were called by the imam of the
mosque […] where a man, named
[A] with a long beard is trying to
reform youth to the only and one
islam…
Example of FCA Lattice
Conceptual time system for a selected subject
Time part
Time granule
Date
Event part
Anti western
Orthodox religion
0
2008-01-26
1
2008-02-15
2
2008-02-24
3
2008-03-28
4
2008-04-06
X
5
2008-01-06
X
6
2008-06-14
7
2008-06-18
Change
behavior
X
X
X
X
X
X
X
X
X
X
TCA lattice of subject A
Extracting and profiling potential jihadists
Showing the subjects
Profiling subject with TCA
Combining business process & data discovery techniques for
analyzing and improving integrated care pathways
• gaining insight in what happens in healthcare process for a
group of patients with same diagnosis
• goal: extract monitor and improve real processes by
extracting knowledge from event logs.
• few studies on process mining in healthcare
• mostly Petri Nets.
Integrated care pathway (ICP): Definition
•
•
•
•
•
Structured multidisciplinary care plan
Details essential steps in the care process
To achieve well defined goals
For a patient with a specific clinical problem
Description of the expected progress over a certain time
period
ICP’s in GZA hospitals
Increasing
Management:
•Provision of high quality care
•On the right moment
•In the optimal circumstances
•By the most appropriate care
provider
•To achieve a predefined result
•At the best cost
•Activity in the organization
•Multidisciplinary care
•Complexity of care
•Patient expectation
Reduction
•Length of stay
•Resources
Case Study: Breast Unit St-Augustinus
Evolution number of patients with
primary operable breast cancer
2002: 194
2003: 256
2004: 245
2005: 385
2006: 391
2007: 379
2008: 381
Complex care process
Phases :
–
–
–
–
Diagnosis
Surgery
Adjuvant treatment
Follow-up
Multidisciplinary team
–
–
–
34 specialists
52 nurses
14 paramedics
Increased need for :
Organizational coordination
Communication
Continuity of care
Healthcare 1.0: Pitfalls
Traditional Business Process Model:
• Do not capture process variations, process exceptions or
root causes of the exceptions and variations
• Only delta analyses to find impact on key performance
indicators
• Models the most standard frequent pathway
• High workload for care process manager
• Retrospective analysis (every 6 months)
• Evaluation over the first semester of each year only
Innovation: Process Discovery
Bottom up approach:
• To discover process inefficiencies, exceptions and variations
immediately
• To gain sufficient understanding of the existing process and
its outliers
• Evaluation of the care process of all the patients
• To search for the root causes of inefficiencies or
improvements
Input for Clinical Path Discovery
• Patient treatment records
• Turned into event sequences
• Compliant with HL7
Expected Benefits of the research
•
•
•
•
•
•
•
Possibility of what if analysis
Reduced waiting times
Decreased average length of stay
Time interval between the interventions
Evaluation immediately available
Decreasing workload of the ICP coordinator
High quality of care with optimal use of the resources
What is Process Discovery about ?
C/K-Theory: Innovative Discovery
C
D is junction:
C onceptualis ation,
T agging & Mining
K
T he C /K
D es ign
S quare
D is covery &
E xploration
C
C onjunction:
Activate & E xperiment
Validation &
L earning
K
74
What is Data Discovery about ?
Combination of process & data discovery (1)
• to gain deeper understanding of existing breast cancer care
process & actual activities performed on work floor
• discover process inefficiencies, exceptions and variations
immediately
• search for root causes of inefficiencies
Combination of process & data discovery (2)
• Hidden Markov Models to discover process models from
event sequences.
• Formal Concept Analysis:
– analyze characteristics of clusters of patients that emerged from
process discovery.
– find groups of patients to feed into the process discovery
methods.
Previous process mining research in healthcare
• mostly Petri-Net models
• Example: process models were built from simulated process
logs of hospital-wide workflows containing events like "blood
test" or “surgery”
• Hidden Markov Model approach: model workflow inside
Operation Room
Hidden Markov Model
• Probabilistic model with greater degree of flexibility
• Better option for healthcare where traditional process mining
does not work well,
• Many (open source) algorithms have been published for
analyzing and understanding HMMs
• Micro patterns of actor behavior can be easily aggregated in
one state
• HMMs can be annotated with a variety of attributes such as
probabilities, time duration, variances, etc.
Dataset
• 148 breast cancer patients hospitalized from January 2008
till June 2008.
• 469 activity identifiers in total
• care trajectory Primary Operable Breast Cancer
• breast cancer care process: 4 phases, 34 doctors, 52 nurses
and 14 paramedics
Breast cancer care process (1)
• every activity performed to a patient is logged in a database
• dataset includes all activities performed during surgery
support phase
Breast cancer care process (2)
Breast cancer data
• each activity has unique identifier
• timestamps assigned to performed activities
• data was turned for each patient into a sequence of events
• sequences of events were input for process discovery
methods
• activities with a similar semantic meaning were clustered to
reduce complexity of lattices and process models
Analysis method
• process models: extraction and visualization of most frequent
standard care pathway
• during analysis of these models: anomalies and process
exceptions are found
• FCA: zoom in on and analyze these observations in detail.
Quality of care analysis
• initial process model: 148 patients and 469 activity codes
• length of stay in hospital < 10 days: linear process
• length of stay > 9 days: 12 patients for which process was
very complex
FCA analysis of 12 patients (1)
• pain score reaches highest point on day 1 and 4 of
hospitalization.
• FCA lattice: overlooked connection between removal of
wound drains and insufficient pain medication.
• pain medication should be administered before removing the
drains
FCA analysis of 12 patients (2)
FCA analysis of 12 patients (3)
• main reason of increased length of stay: neurological /
psychiatric problems, wound infection, subsequent bleeding.
• cancer care process more complex resulting in more
investigative tests.
• since additional morbidities are a root cause for this
increased length of stay: treatment should be anticipated on
& optimalized during preoperative phase
Process variations (1)
5 types of breast cancer surgery:
–
–
–
–
–
mastectomy
breast conserving surgery
lymph node removal
combination of mastectomy and lymph node removal
combination of breast conserving surgery and lymph node
removal
Process variations (2)
• For each surgery type:
– process model was built
– FCA lattice for analyzing characteristics of patient groups
• Mastectomy vs. breast conserving surgery
– more complex surgery type
– FCA lattices were less complex for mastectomy than for breast
conserving surgery.
Breast conserving surgery (1)
Breast conserving surgery (2)
• less uniformly structured care process
• essential care interventions are missing
– 3 patients did not receive consultation from social support
service
– 15 patients did not have appointment with physiotherapist & did
not receive revalidation therapy.
– 1 patient did not receive pre- operative preparation
– 2 patients were missing emotional support before and after
surgery
Breast conserving surgery (3)
• original pathway was written for certain length of stay
• length of stay was significantly reduced over past years
without modifying the care process model
• became impossible to execute prescribed process model in
practice
• patients are receiving suboptimal care
Breast conserving surgery (4)
• Solution:
– activities performed to patients should be reorganized
– care pathway taking into account this time restriction should be
optimalized
Mastectomy (1)
Mastectomy (2)
• less complex lattice structure although care is more complex
• most patients received all key intervention prescribed in
clinical pathway
• 2 patients with quality of care issue:
– 1 patient did not receive emotional support
– 1 patient did not receive a breast prosthesis before discharge
Workforce intelligence (1)
Workforce intelligence (2)
• 25 patients with LOS < 4 days are treated by surgeon 9.
• patients treated by other doctors have longer LOS
• process models were constructed for patients with
– LOS smaller than 4 days
– LOS equal to 4 days
– LOS larger than 4 days.
• extract best practices.
Process models
ML LOW
ML AVG
ML HIGH
Data entrance quality problems
• some patients for who activities were registered after day of
discharge
• reason: error in computer program combined with sloppy
data entry by nursing staff
• semantically identical activities that had different activity
numbers
Data entrance quality problems (2)
• process models have ordering of events that does not
correspond to ordering in real life
– reason: error in computer system which sometimes imposes
certain sequences of events
• discrepancy between built-in top-down developed model and
reality
– reason: insufficient insight into reality of working floor
Journal Articles: Published
Poelmans, J., Elzinga, P., Viaene, S., Van Hulle, M. & Dedene G. (2009). Gaining insight in domestic
violence with emergent self organizing maps, Expert systems with applications, 36, (9), 11864 –
11874. [SCI=2.596]
Poelmans, J., Elzinga, P., Viaene, S., Dedene, G., Van Hulle, M. (2009). Analyzing domestic violence with
topographic maps: a comparative study, Lecture Notes in Computer Science, 5629, 246 – 254,
Advances in Self-organizing Maps, 7th International Workshop on Self-Organizing Maps (WSOM). St.
Augustine, Florida (USA), 8-10 June 2009, Springer. [SCI=0.295]
Poelmans, J., Elzinga, P., Viaene, S., Dedene, G. (2009). A case of using formal concept analysis in
combination with emergent self organizing maps for detecting domestic violence, Lecture Notes in
Computer Science, 5633, 247 – 260, Advances in Data Mining. Applications and Theoretical Aspects,
9th Industrial Conference (ICDM), Leipzig, Germany, July 20-22, 2009, Springer. [SCI=0.295]
Poelmans, J., Elzinga, P., Viaene, S., Dedene, G. (2008). An exploration into the power of formal concept
analysis for domestic violence analysis, Lecture Notes in Computer Science, 5077, 404 – 416,
Advances in Data Mining. Applications and Theoretical Aspects, 8th Industrial Conference (ICDM),
Leipzig, Germany, July 16-18, 2008, Springer. [SCI=0.295]
Journal Articles: Published
Poelmans, J., Elzinga, P., Viaene, S., Dedene, G. (2010), Formal Concept Analysis in knowledge
discovery: a survey. Lecture Notes in Computer Science, 6208, 139-153, 18th international conference
on conceptual structures (ICCS 2010): from information to intelligence. 26 - 30 July, Kuching,
Sarawak, Malaysia. Springer.
Manyakov, N., Poelmans, J., Vogels, R., Van Hulle, M. (2010), Combining ESOMs trained on hierarchy of
feature subsets for single-trial decoding of LFP responses in monkey area V4. Lecture Notes in
Artificial Intelligence, 6114, 548-555, 10th International Conference on Artificial Intelligence and Soft
Computing. June 13-17, Zakopane, Poland. Springer
Poelmans, J., Dedene, G., Verheyden, G., Van der Mussele, H., Viaene, S., Peters, E. (2010). Combining
business process and data discovery techniques for analyzing and improving integrated care
pathways. Lecture Notes in Computer Science, Advances in Data Mining. Applications and Theoretical
Aspects, 10th Industrial Conference (ICDM), Leipzig, Germany, July 12-14, 2010. Springer
Vuylsteke A., Baesens B., Poelmans J. (2010). Consumers’ search for information on the internet: how and
why China differs from Western Europe, Accepted for Journal of interactive marketing.
Journal Articles: Submitted
Verheyden, G., Poelmans, J., Viaene, S., Van der Mussele, H., Dedene, G., van Dam, P. (2010). Key
Success Factors for significantly improving Patient Satisfaction on Breast Cancer care: a Case Study,
submitted for The Breast
Poelmans, J., Elzinga, P., Viaene, S., Dedene, G. (2010) Curbing domestic violence: Instantiating C-K
theory with Formal Concept Analysis and Emergent Self Organizing Maps, submitted for IEEE
transactions on knowledge and data engineering.
Poelmans, J., Elzinga, P., Viaene, S., Van Hulle, M. & Dedene G. (2010) Text Mining with Emergent Self
Organizing Maps and Multi-Dimensional Scaling: A comparitive study on domestic violence, submitted
for Applied Soft Computing.
Poelmans, J., Elzinga, P., Viaene, S., Dedene, G. (2010) Formal Concept Analysis in Information
Engineering: a Survey, submitted for ACM Computing Surveys.
Poelmans, J., Elzinga, P., Viaene, S., Dedene, G. (2010) Formally Analyzing the Concepts of Domestic
Violence, submitted for Expert Systems with Applications.
Journal Articles: Submitted
Poelmans, J., Elzinga, P., Viaene, S., Dedene, G. (2010) Informatiegestuurd handhaven: een slimme kijk
naar bestaande data, submitted for Informatie.
Elzinga, P., Poelmans, J., Viaene, S., Dedene, G. (2010) Formele concept analyse: een nieuwe dimensie
voor intelligence, submitted for Blauw.
Conference proceedings: Accepted
Poelmans, J., Dedene, G., Snoeck, M. Viaene, S. (2010). Using Formal Concept Analysis for the
Verification of Process-Data matrices in Conceptual Domain Models, Proc. IASTED International
Conference on Software Engineering (SE 2010), Feb 16 - 18, Innsbruck, Austria. Acta Press, pp..
Poelmans, J., Elzinga, P., Viaene, S., Dedene, G. (2010). A method based on Temporal Concept Analysis
for detecting and profiling human trafficking suspects. Proc. IASTED International Conference on
Artificial Intelligence (AIA 2010). Innsbruck, Austria, 15-17 february. Acta Press ISBN 978-0788986817-5, pp. 330-338.
Elzinga, P., Poelmans, J., Viaene, S., Dedene, G. (2009), Detecting domestic violence – Showcasing a
Knowledge Browser based on Formal Concept Analysis and Emergent Self Organizing Maps, Proc.
11th International Conference on Enterprise Information Systems ICEIS, Volume AIDSS, pp. 11 – 18,
Milan, Italy, May 6-10, 2009.
Poelmans, J, Elzinga, P., Van Hulle, M., Viaene, S., and Dedene, G. (2009). How Emergent Self
Organizing Maps can help counter domestic violence, World Congress on Computer Science and
Information Engineering (CSIE 2009), Los Angeles (USA), Vol. 4, IEEE Computer Society Press ISBN
978-0-7695-3507-4, 126 – 136.
Conference proceedings: Accepted
Vuylsteke, A., Wen, Z., Baesens, B. and Poelmans J. (2009). Consumers Online Information Search: A
Cross-Cultural Study between China and Western Europe. Paper presented at Academic And
Business Research Institute Conference 2009, Orlando, USA, available at
http://www.aabri.com/OC09manuscripts/OC09043.pdf
Elzinga, P., Poelmans, J., Viaene, S., Dedene, G., Morsing, S. (2010) Terrorist threat assessment with
Formal Concept Analysis. Proc. IEEE International Conference on Intelligence and Security
Informatics. May 23-26, 2010 Vancouver, Canada. ISBN 978-1-42446460-9/10, 77-82.
Conference proceeding, book chapter & Dutch
publications
Dejaeger, K., Hamers, B., Poelmans, J., Baesens, B. (2010) A novel approach to the evaluation and
improvement of data quality in the financial sector, submitted for 15th International Conference on
Information Quality (ICIQ 2010) UALR, Little Rock, Arkansas USA,
Poelmans J, Van Hulle M, Elzinga P, Viaene S, Dedene G (2008) Topographic maps for domestic violence
analysis. Self-organizing maps and the related tools, pp. 136 - 145.
Vuylsteke A., Poelmans J., Baesens B. (2009) Online zoekgedrag van consumenten: China vs WestEuropa, Business In-Zicht, December, 2-3.
Dejaeger K., Ruelens J., Van Gestel T., Jacobs J., Baesens B., Poelmans J., Hamers B. (2009) Evaluatie
en verbetering van de datakwaliteit. Informatie, November , Jaargang 51/9, 8-15.
Awards
Nominated for best paper award at 8th Industrial Conference on Data Mining
(ICDM), Leipzig, Germany, July 16-18, 2008
Winner of young professionals best paper award at 9th Industrial Conference on
Data Mining (ICDM), Leipzig, Germany, July 20-22, 2009
Winner of best paper award at 10th Industrial Conference on Data Mining
(ICDM), Berlin, Germany, July 12-14, 2010