DETECTING DOMESTIC VIOLENCE

Showcasing a Knowledge Browser based on Formal Concept Analysis and

Emergent Self Organizing Maps

Paul Elzinga

Department of Business Information, Police Amsterdam-Amstelland, The Netherlands

Jonas Poelmans, Stijn Viaene, Guido Dedene

Faculty of Business and Economics, K.U. Leuven, Belgium

Vlerick Leuven Gent Management School, Belgium

Keywords: Formal Concept Analysis (FCA), Emergent SOM, Domestic violence, Knowledge discovery in databases,

Text mining, Exploratory data analysis.

Abstract: Over 90% of the case data from police inquiries is stored as unstructured text in police databases. We use

the combination of Formal Concept Analysis and Emergent Self Organizing Maps for exploring a dataset of

unstructured police reports out of the Amsterdam-Amstelland police region in the Netherlands. In this

paper, we specifically aim at making the reader familiar with how we used these two tools for browsing the

dataset and how we discovered useful patterns for labelling cases as domestic or as non-domestic violence.

1 INTRODUCTION

Over 90% of the information available to police

organizations is stored as plain text. To date,

however, analyses have primarily focused on the

structured portion of the available data. Only

recently the first steps for applying text mining in

criminal analysis have been taken (Ananyan, 2002;

Chen, 2004). In the case we are studying, we aim at

automating the detection of domestic violence using

the unstructured text contained in police reports. We

use real-life data recorded during 2007 by the

Amsterdam-Amstelland police in The Netherlands.

In 1997, the ministry of Justice of the

Netherlands made its first inquiry into the nature and

scope of domestic violence. It turned out that 45% of

the population once fell victim to non-incidental

domestic violence. For 27% of the population, the

incidents even occurred on a weekly or daily basis.

These gloomy statistics placed the topic high on the

political agenda.

Pursuing an effective policy against offenders is

one of the top priorities of the regional police

Amsterdam-Amstelland. In order to do this, being

able to swiftly recognize cases of domestic violence

and label reports accordingly is of the utmost

importance. Immediately after the reporting of a

crime, police officers are given the possibility to

judge whether or not it is a domestic violence case.

If they believe it is, they can assign the label

domestic violence to the report. That, however, has

proven problematic. In the past, intensive audits of

the police databases related to filed reports have

shown that many reports are not well classified.

Therefore, a case triage system has been put in

place to automatically filter out suspicious cases for

in-depth manual inspection and classification. Still,

to date more than 80% of these suspicious cases are

wrongly selected for in-depth inspection by the

system (i.e. false positives). Given that it takes at

least five minutes to read and classify a case, it is

clear that building an efficient and automated case

labelling system would result in major savings.

In 2007, the database of the Amsterdam-

Amstelland police contained more than 7000 cases

with a statement made by the victim of a violent

incident to the police. Because it is physically

impossible for any individual to process this sheer

amount of information, applying text mining

technology seems a natural approach. Text mining

has been defined as “the discovery by computer of

Elzinga P., Poelmans J., Viaene S. and Dedene G. (2009).

DETECTING DOMESTIC VIOLENCE - Showcasing a Knowledge Browser based on Formal Concept Analysis and Emergent Self Organizing Maps.

In Proceedings of the 11th International Conference on Enterprise Information Systems - Artiﬁcial Intelligence and Decision Support Systems, pages

11-18

DOI: 10.5220/0001844400110018

 SciTePress

new, previously unknown, information by

automatically extracting information from different

written resources” (Fan, 2006).

For our data exploration we chose to use a

combination of two visually appealing discovery

techniques, known as Formal Concept Analysis

(FCA) (Stumme, 2002) and Emergent Self

Organizing Maps (ESOM) (Ultsch, 2005). Formal

Concept Analysis (FCA) arose twenty-five years ago

as a mathematical theory (Stumme, 2002). FCA was

for the first time used as an exploratory data analysis

and knowledge enrichment technique for analysing

domestic violence cases in the Netherlands

(Poelmans, 2008). In this setup, FCA is used as a

concept generation engine, distilling formal concepts

from the unstructured documents. We complement

the knowledge discovery based on FCA with the

ESOM. Emergent Self Organizing Maps are a

special class of topographic maps, which are

particularly suited for high-dimensional data

visualization.

In this paper we aim at making the reader

familiar with how we used these tools for browsing

through the data in search of new knowledge for

classifying new cases. The result of the research is a

case labelling system that automatically and

correctly assigns the domestic violence or non-

domestic violence label to a large portion of the

incoming cases.

The rest of this paper is structured as follows. In

section 2, the dataset is discussed. In section 3, the

use of FCA and ESOM for knowledge discovery is

presented and applied to the data at hand. In section

4, the ensuing detection process is summarized.

Section 5 concludes the paper.

2 THE DATASET

According to the Dutch police authorities and the

department of Justice, domestic violence can be

characterized as serious acts of violence committed

by someone of the domestic sphere of the victim.

Violence includes all forms of physical assault. The

domestic sphere includes all partners, ex-partners,

family members, relatives and family friends of the

victim. Family friends are those persons who have a

friendly relationship with the victim and who

(regularly) meet the victim in his/her home (Van

Dijk, 1997).

The dataset we report on in this paper consists

of a selection of 4814 police reports describing a

whole range of violent incidents from the year 2007.

The domestic violence cases for that period are a

subset of this dataset. This selection came about by,

among other things, filtering from a larger set those

police reports that did not contain the reporting of a

crime by a victim, which is necessary for

establishing domestic violence. This happens, for

example, when a police officer is sent to an incident

and later on writes a report in which he/she mentions

his/her findings, while the victim has not made an

official statement to the police. The follow-up

reports referring to previous cases were also

removed from the initial set of reports. Ultimately,

this gave rise to a set of 4814 reports that were used

as input for our investigation. From these reports,

the person who reported the crime, the suspect, the

persons involved in the crime, the witnesses, the

project code and the statement made by the victim to

the police were extracted. Of the 4814 reports, 1657

were classified by police officers as domestic

violence; the others were not.

3 HUMAN-CENTERED

KNOWLEDGE DISCOVERY

In the literature, the need for exploratory data

analysis has often been described (Marchionini,

2006). When beginning the analysis of a new dataset

of which very little is known a priori, the first step is

to explore the data. Data mining should be primarily

concerned with making it easy, convenient and

practical to explore very large databases for

organizations with a lot of users but without

requiring years of training into data analysis

(Fayyad, 2002). Unfortunately, much attention and

effort has been focused on the development of data

mining techniques but only a minor effort has been

devoted to the development of tools that support the

analyst in the overall discovery task (Brachman,

1996). A human-centered approach is proposed. A

significant part of the art of data mining is the user’s

intuition with respect to the tools (Smyth, 2002). We

argue that the combined use of FCA and ESOM

fulfils this need. The visual representations of both

tools provide a clear guide to the user for exploring

the data.

Additionally, we aim at developing a classifier

for automatically classifying cases as domestic or as

non-domestic violence. Comprehensibility of the

performed classification is a key requirement,

requiring that the user understands the motivations

behind the model’s prediction (Martens, 2004). In

the domain of police investigations, the lack of

comprehensibility is a major issue and causes a

ICEIS 2009 - International Conference on Enterprise Information Systems

reluctance to use a classifier or even complete

rejection of the model. This is for a large part due to

the very high cost of classifying a case incorrectly as

domestic or as non-domestic violence. Clarity and

explainability of the performed classification are

major constraints. Comprehensibility measures the

“mental fit” of the classification model (Kodratoff,

1994).

The knowledge discovery process using the

combination of FCA and ESOM is displayed in

Figure 1.

Figure 1: Knowledge acquisition process.

The knowledge acquisition process is explained

and showcased in the next subsections.

3.1 Data Preparation

The initial phase of the knowledge acquisition

process consists of translating the investigation area

into objects, terms and attributes. We considered the

police reports from the dataset as objects and the

relevant terms contained in these reports as

attributes. The terms and term clusters (see section

3.2) are stored in a thesaurus.

We composed an initial thesaurus of which the

content was based on expert prior knowledge such

as the domestic violence definition. We enriched the

thesaurus with terms referring to the different

components of the definition such as “hit”, “stab”,

“my mother”, “my ex-boyfriend”, etc. Since

according to the literature domestic violence is a

phenomenon that typically occurs inside the house,

we also added terms such as “bathroom”, “living

room”, etc. We made an explicit distinction with

public locations such as “under the bridge”, “on the

street”, etc. The initial thesaurus contained 123

elements. An excerpt of this thesaurus is shown in

Figure 2.

Figure 2: Thesaurus.

The reports were indexed using this thesaurus.

For each report the found thesaurus elements were

stored in a collection. This collection would be used

as an input for both the FCA and the ESOM

processing.

The thesaurus was refined after each iteration of

re-indexing the reports and visualizing and analysing

the data with the FCA lattice and ESOM maps. This

process is showcased in section 3.2 and 3.3.

3.2 Exploratory Analysis with FCA

The tool we used to visualize and analyze the data

with FCA is the tool Conexp (Yevtushenko, 2000).

The starting point of the FCA analysis is a

database table consisting of rows (i.e. objects),

columns (i.e. attributes) and crosses (i.e.

relationships between objects and attributes). The

mathematical structure used to reference such a

cross table is called a formal context. An example of

a cross table is displayed in Figure 3. In the latter,

reports of domestic violence (i.e. the objects) are

related (i.e. the crosses) to a number of terms (i.e.

the attributes); here a report is related to a term if the

report contains this term.

Prior to the analysis with FCA, the terms in the

thesaurus have to be clustered in term clusters. The

initial clustering, performed on the basis of expert

prior knowledge, resulted in four term clusters: “acts

DETECTING DOMESTIC VIOLENCE - Showcasing a Knowledge Browser based on Formal Concept Analysis and

Emergent Self Organizing Maps

of violence”, “public locations”, “private locations”

and “persons”.

The formal context based on the domestic

violence definition is displayed in Figure 3.

Figure 3: Initial cross table.

FCA uses the mathematical abstraction of the

concept lattice to describe systems of concepts to

support human actors in their information discovery

and knowledge creation exercise.

Given a formal context, FCA derives all concepts

from this context and orders them according to a

subconcept-superconcept relation. This results in a

line diagram, a.k.a. lattice. The line diagram

corresponding to the cross table from Figure 3 is

represented in Figure 4.

Figure 4: Initial lattice.

When the lattice has been built, it can be

analyzed using the Conexp tool. The tool provides a

user-friendly interface to navigating the lattice. The

user can highlight specific aspects of the lattice,

select and/or deselect attributes, display the case

numbers that belong to each of the concepts, etc.

Figure 5: Lattice with own object count.

For example, the lattice displayed in Figure 5

was obtained by deselecting the attributes “private

locations” and “public locations,” and by showing

the number of own objects for each concept.

Another example is the lattice displayed in

Figure 6 that was obtained by making the case

numbers belonging to an interesting concept visible.

Figure 6: Lattice with case numbers.

ICEIS 2009 - International Conference on Enterprise Information Systems

Some interesting findings that emerged from

navigating the lattice are displayed in Table 1.

Table 1: Key statistics from initial lattice navigation.

Non-domestic

violence

Domestic

violence

No “acts of violence” 67 41

No “acts of violence”

and “persons”

61 19

Only “acts of

violence”

877 64

As can be seen from Table 1, a total of 60 (i.e.

41 and 19) domestic violence cases did not contain a

term from the “acts of violence” term cluster. Of

these 60 cases 19 contained a term from one of the

clusters containing terms referring to a person in the

domestic sphere of the victim. After in-depth manual

inspection of these 19 cases, it turned out that they

contained other violence terms, such as “abduction”,

“strangle” and “deprivation of liberty”, which were

lacking in the initial thesaurus. The remaining 41

cases, on the other hand, turned out to be wrongly

classified as domestic violence.

Interestingly, some 28% (i.e. 877) of the non-

domestic violence reports only contain terms from

the “acts of violence” cluster, while there are only

64 domestic violence reports in the dataset that share

that characteristic. Manual inspection, again,

revealed that more than two thirds of these reports

were wrongly classified as domestic violence. For

some unknown reason, police officers regularly

seem to misclassify burglary, car theft, bicycle theft

and street robbery cases as domestic violence.

Therefore, terms such as “street robbery”, burglary”

and “car theft” were combined into a new term

cluster called “burglary cases”. These term clusters

are used as classification rules. If a case contains a

term from the “burglary cases” cluster, for example,

we found that it can be correctly classified as non-

domestic violence.

After several iterations of exploring and refining

the thesaurus, 44% of the cases could automatically

be classified using such classification rules. To

complement the use of FCA, the ESOM was

introduced to function as a catalyst during the

knowledge acquisition process, amongst others for

finding new terms and term clusters. The ESOM tool

is described in detail in the next section.

3.3 Exploratory Analysis with ESOM

The ESOM performs a non-linear mapping of the

high-dimensional data space to a two-dimensional

one which enables the exploration of the data

(Ultsch, 2003). Emergence is the ability of a system

to produce a phenomenon on a new, higher level

(Ultsch, 1999). In order to achieve emergence, the

existence and cooperation of a large number of

elementary processes is necessary. An Emergent

SOM differs from a traditional SOM in that a very

large number of neurons (at least a few thousands)

are used. In the traditional SOM, the number of

nodes is too small to show emergence.

The ESOM can be used to detect clusters and

maintains the neighbourhood relationships that are

present in the input space. It also provides the user

with an idea of the complexity of the dataset, the

distribution of the dataset and the amount of overlap

between the different classes. Only a minimal

amount of expert knowledge is required for the user

to be able to use it effectively for exploratory data

analysis. The maps can be created and used for data

analysis by means of the publicly available

Databionics ESOM Tool (Ultsch, 2005). With this

tool the user can construct ESOMs with either flat or

unbounded (i.e., toroidal) topologies.

The ESOM is used as a catalyst in the

knowledge discovery process, amongst others for

finding new terms and term clusters. ESOM

provides a highly interactive user interface that

moves exploratory search process beyond

predictable fact retrieval. Because of its highly

interactive user interface, human participation and

effective use of their expert prior knowledge in the

search process is promoted.

Indexing the 4814 reports from 2007 with the

initial thesaurus from section 3.1 resulted in a cross

table with all reports as objects and all terms as

attributes. This cross table is used for training a

toroidal ESOM. The ESOM is presented in Figure 7.

The green squares refer to neurons that dominantly

contain non-domestic violence cases, while the red

squares refer to neurons that dominantly contain

domestic violence cases.

Some red squares in Figure 7 are located in the

middle of a large group of green squares and vice

versa. The ESOM tool allows the user to select

neurons on the map (yellow area in the map of

Figure 7). It displays the cases that had this neuron

as a best match in the lower pane of Figure 7. After

an in-depth manual inspection of the police reports

corresponding to these outliers, interesting

discoveries were made. For example, we observed

that many of these outlier reports contained several

important new features that were lacking in the

domain expert’s understanding of the problem area.

The reports also contained multiple confusing

DETECTING DOMESTIC VIOLENCE - Showcasing a Knowledge Browser based on Formal Concept Analysis and

Emergent Self Organizing Maps

situations that upon disclosure to us were used to

refine the domestic violence definition.

Figure 7: Toroidal ESOM trained on the thesaurus from

section 3.1.

Every time new and important features were

discovered in this way, they were used to enrich the

thesaurus, and triggered a new iteration of the FCA

analysis, followed by another run of the ESOM tool.

In each iteration, it is possible that one or more new

classification rules are discovered. The attribute

“corporate body”, for example, was found by first

analyzing a cluster of green squares that was located

within a group of red squares in a map. With FCA,

we found that the presence of a corporate body in a

police report almost always excludes domestic

violence. Therefore, we introduced a new domestic

violence classification rule named “corporate body”.

4 DOMESTIC VIOLENCE

DETECTION SYSTEM

Based on the knowledge discovery laid out in the

previous sections, we developed a Tomcat-based

system to assist analysts in their labelling of cases.

The system is currently used as a stand-alone

application by the data quality management team

(i.e. the back office). The long term goal is to make

it available to all police officers in the organization

(i.e. the front office) to assist them in their labelling

of cases.

4.1 Classification Rules

After a successive number of knowledge acquisition

iterations using FCA and ESOM, we were able to

enrich the initial thesaurus from 123 to 829

thesaurus elements. These 829 thesaurus elements

include 50 term clusters and 779 domain-specific

terms.

With FCA a set of 22 domestic violence and 15

non-domestic violence classification rules were

extracted. Using these rules, 75% of cases from the

year 2007 can be classified automatically as either

domestic or non-domestic violence. We also applied

these rules to a validation set containing

unstructured police reports from the year 2006 and a

similar result was obtained. In the near future, these

rules can be used to automatically classify the

majority of incoming cases, while in the past all

cases had to be dealt with manually

Ten of the domestic violence and five of the non-

domestic violence classification rules are displayed

in Table 2.

Table 2: Classification rules.

Domestic violence classification rules

1 Legal proceedings against domestic sphere

2 Committed by domestic sphere

3 Relational problems and living together

4 Relational problems and institutions

5 Honour related violence

6 Incest

7 (Court) injunction

8 Fear of domestic sphere

9 Committed by domestic sphere

10 Problems with domestic sphere

Non-domestic violence classification rules

1 Unknown perpetrator

2 Corporate body

3 Burglary cases

4 Road rage

5 Violence at school

These classification rules were also used to

detect filed reports that were assigned a faulty case

label.

In the next section we discuss the application

that was developed to apply the discovered

knowledge to classify police reports.

4.2 Labelling Process

The labelling process, as performed by the data

quality management team, consists of a number of

steps that are, to a large extent, automated by the

newly introduced system. First, the user can select a

ICEIS 2009 - International Conference on Enterprise Information Systems

set of police reports for labelling (e.g. all police

reports from the month October 2008). A standard

thesaurus is provided, but the user can modify or

switch this thesaurus at will. The next step consists

of indexing the set of police reports using the

selected thesaurus and generating a cross table. The

text mining tool Collexis is used to index the police

reports and translate the found attributes into Prolog-

facts.

Subsequently, the classification rules that were

discovered during the exploration of the data are

applied to the cases. When a case comes in for

labelling, the first step consists in verifying whether

one of the domestic violence rules is satisfied. If this

is the case, the case is classified as domestic

violence. Otherwise, it is verified whether one of the

non-domestic violence rules is satisfied. If this is the

case, the case is classified as non-domestic violence.

Otherwise, the case is left unclassified.

The rule base was developed in Prolog, in order

to make the knowledge base more flexible and

expandable for other areas such as discrimination

and weapons. The result of applying the

classification rules to a set of police reports is

displayed in Figure 8.

Figure 8: Domestic violence detection system.

For each report, the assigned label and the

applied rule are shown. Moreover, for each case,

there is a hyperlink on which the user can click to

open the corresponding report in a popup window.

As shown in Figure 9, the terms from the thesaurus,

that were found in the report, are highlighted.

Figure 9: Police report with highlighted thesaurus terms.

The user can quickly verify whether the

classification is correct. The application will also

highlight the found attributes if no classification can

be made (in 25% of the cases). The highlighting of

terms makes it easier for the user to label a case.

5 CONCLUSIONS

This paper showed how the Amsterdam-Amstelland

police is using text mining on police reports to better

identify cases of domestic violence. At the heart of

the knowledge discovery process are the lattices

produced with FCA and the maps produced with

ESOM. The knowledge browsing capabilities of

these tools provide for a powerful framework for

exploring unstructured data. In our search for

improved detection of domestic violence cases, we

discovered 37 classification rules, i.e. 22 domestic

violence and 15 non-domestic violence rules. With

these rules 75% of cases can be labelled

automatically and correctly, whereas in the past, all

cases had to be dealt with manually. These rules are

the motor of a detection system that aids analysts in

their assessment of incoming cases.

ACKNOWLEDGEMENTS

The authors would like to thank the police of

Amsterdam-Amstelland for providing them with the

necessary degrees of freedom to conduct and publish

DETECTING DOMESTIC VIOLENCE - Showcasing a Knowledge Browser based on Formal Concept Analysis and

Emergent Self Organizing Maps

this research. In particular, we are most grateful to

Deputy Police Chief Reinder Doeleman and Police

Chief Hans Schönfeld for their continued support.

Jonas Poelmans is Aspirant of the “Fonds voor

Wetenschappelijk Onderzoek – Vlaanderen” (FWO)

or Research Foundation – Flanders.

REFERENCES

Brachman, R., Anand, T., 1996. The process of knowledge

discovery in databases: a human-centered approach.

In advances in knowledge discovery and data mining,

ed. U. Fayyad, G. Piatetsky-Shapiro, P. Smyth and R.

Uthurusamy. AAAI/MIT Press.

Chen, H., Chung, W., Xu, J.J., Wang, G., Qin, Y., Chau,

M., 2004. Crime data mining: a general frame-work

and some examples. IEEE Computer, April 2004.ijk, T

van, 1997. Huiselijk geweld, aard, omvang en hulp-

verlening. Ministerie van Justitie, Dienst Preventie,

Jeugd-bescherming en Reclassering.

Fan, W., Wallace, L., Rich, S., Thang, T., 2006. Tapping

the power of text mining. Communications of the

ACM, Vol. 49, no. 9.

Fayyad, U., Uthurusamy, R., 2002. Evolving data mining

into solutions for insights. Communications of the

ACM, Vol. 45, no. 8.

Kodratoff, Y., 1994. The comprehensibility manifesto.

KDD nuggets (94:9).

Marchionini, G., 2006. Exploratory search: from finding

to understanding. Communications of the ACM, Vol.

49, no. 4.

Martens, D., Baesens, B., Van Gestel, T., Vanthienen, J.,

2007. Comprehensible credit scoring models using

rule extraction from support vector machines.

European journal of operational research, 183(3),

1466-1476.

Poelmans, J, Elzinga P, Viaene S., Dedene G, 2008. An

exploration into the power of Formal Concept

Analysis for domestic violence analysis. 8

Industrial

Conference, ICDM 2008 Leipzig, Germany, Juli 2008

Proceedings, LNAI 5077 Springer Heidelberg, pp.

404-416.

Pednault, E.P.D. 2000, Representation is everything.

Communications of the ACM, Vol. 43, no. 8.

Smyth, P., Pregibon, D., Faloutsos, C., 2002. Data-driven

evolution of data mining algorithms. Communications

of the ACM, Vol. 45, no. 8.

Stumme, G., 2002. Formal Concept Analysis on its Way

from Mathematics to Computer Science. Proc. 10

Intl. Conf. on Conceptual Structures (ICCS 2002).

LNCS, Springer, Heidelberg 2002.

Ultsch, A., 1999. Data mining and knowledge discovery

with Emergent SOFMS for multivariate Time Series.

Kohonen Maps, pp. 33-46.

Ultsch, A., 2003. Maps for visualization of high-

dimensional Data Spaces. In proc. WSOM’03,

Kyushu, Japan, pp. 225-230.

Ultsch, A., Moerchen, F., 2005. ESOM-Maps: Tools for

clustering, visualization, and classification with

Emergent SOM. Technical Report Dept. of

Mathematics and Computer Science, University of

Marburg, Germany, No. 46.

Yevtushenko, S.A., 2000. System of data analysis

“Concept Explorer.” Proceedings of the 7

national

conference or Artificial Intelligence. KII-2000. 127-

134, Russia.

ICEIS 2009 - International Conference on Enterprise Information Systems