VISUAL TREND ANALYSIS

Ontology Evolution Support for the Trend Related Industry Sector

Jessica Huster

1,2

and Andreas Becks

Fraunhofer-Institute for Applied Information Technology FIT, Schloss Birlinghoven, 53754 Sankt Augustin, Germany

Informatik 5 (Information Systems), RWTH Aachen University Ahornstr. 55, 52056 Aachen, Germany

Keywords: Text Mining, Concept-drifts, Ontology Evolution.

Abstract: Ontologies are used as knowledge bases to exchange, extract and integrate information in information

retrieval and search. They provide a shared and common understanding that reaches across people and

application systems. In reality, domain specific and technical knowledge evolve over time, and so must

ontologies. Creative domains, as for example the home textile industry, are representatives for quickly

evolving domains. In this domain it is also important to provide methodologies for the visualisation of

knowledge evolution. In this paper we report on our ontology-based trend analysis tool, which supports

marketing experts and designers to identify trend drifts, and to compare the analysis results against the

ontology. Furthermore means to adapt and evolve the ontology in accordance with the changing domain are

provided.

1 INTRODUCTION

Ontologies have been proven to be useful in

different fields such as information integration,

search, and retrieval. Regarding the Semantic Web,

ontologies with their definition of concepts and

relationships provide the backbone for structured

access and exchange of shared knowledge (Fensel,

2001b), (Flouris et al., 2008).

Hence ontologies are seen as and used for “a

shared and common understanding that reaches

across people and application systems” (Fensel,

2001a). However, different definitions and broad

interpretations exist about what can be referred to as

ontology. These interpretations include simple

catalogues, sets of structured text files, thesauri,

taxonomies and also sets of general logical

constraints, that enable automated reasoning (Welty

et al., 1999). The latter models usually have a

greater complexity than the former. The ontology

used in this paper is a lightweight, terminological

ontology that describes the domain of the home

textile industry, i.e. curtains, furnishing fabrics and

carpets. It mainly uses mainly sub-class-

relationships.

Ontologies represent the current understanding

of terms and their relations of a domain in a

structured way, ensuring that a group of people

agree upon the same meaning of terms. However,

ontologies, once created, do not last forever because

they do not only encapsulate stable knowledge. In

reality, the domains which are modelled by

ontologies evolve, and so must the ontologies to stay

useful (Fensel, 2001b), (Stojanovic et al., 2002),

(Haase and Sure, 2004). There is a need for dynamic

evolution in the conceptualisation to ensure

reliability and effective support through ontologies.

Meanwhile, the issue of adapting ontologies due

to a certain need of change is addressed by several

different, but also closely related research

disciplines. This field is summarised under the term

of ontology change and comprises several subfields,

each focusing on another aspect of ontology change.

Flouris et al. (2008) identify 10 subfields in their

work, namely: ontology mapping, morphism,

matching, articulation, translation, evolution,

debugging, versioning, integration and merging.

Regarding the Semantic Web the field of ontology

evolution is of particular importance as distributed,

heterogonous and dynamic domains represented by

different ontologies are expected to cooperate in this

field.

The term ontology evolution is used with

different meanings in the literature. In this paper, we

will adapt the definition by Flouris et al. (2007):

ontology evolution is the “process of modifying an

Huster J. and Becks A. (2010).

VISUAL TREND ANALYSIS - Ontology Evolution Support for the Trend Related Industry Sector.

In Proceedings of the 12th International Conference on Enterprise Information Systems - Artiﬁcial Intelligence and Decision Support Systems, pages

46-53

 SciTePress

ontology in response to a certain change in the

domain or its conceptualization”. Thus, our focus is

on the process of adaptation of an ontology. We

distinguish this process from ontology versioning

which deals with maintaining several different, but

related versions of the same ontology.

Creative domains are one example for quickly

evolving domains in which new concepts and

relationships are very frequently established. As an

example for a creative domain in this paper, we will

consider the domain of home textile industry.

Stakeholders in this domain have to flexibly adjust

to upcoming and descending trends. Therefore, the

issue of modifying a domain ontology is important

in this context. Furthermore, in such a domain it is

important to support the modification of the domain

ontology by appropriate change and update

operations. Additionally, methodologies for the

visualisation of the knowledge evolution are

required as users need to be able to understand how

the context and importance of certain concepts in the

domain have changed. In this paper we present an

ontology-based system that supports this industry

sector to monitor trend relevant sources and identify

drifts in their market. Hence, we set a focus on an

intuitive change capturing, which forms the starting

point for ontology evolution. Once the changes,

which require a modification of the ontology have

been identified, the system provides means to adapt

the ontology accordingly.

The remaining paper is structured as follows:

The next section presents an overview on existing

approaches to the topic of evolving and maintaining

ontologies. Section 3 presents our approach in more

detail. Section 3.1 covers the difficulty in the

creative domain of home textile sector followed by

the system description (section 3.2). Section 4

presents the evaluation conducted in cooperation

with carpet producing companies.

2 RELATED WORK

Many approaches focused on creating ontologies

rather than on ontology evolution. This shows that

these knowledge models have been treated mainly as

static and that the encapsulated domain knowledge is

assumed to not change (Haase and Sure, 2004).

The approaches of ontology creation and

evolution can be distinguished into two main groups.

One class of evolution methods follows a

community based approach, whereas the other class

of approaches tries to automate, at least part of the

process and suggest computational methods for

ontology creation.

An early work in this area to construct ontologies

collaboratively is presented by Dominge (1998). His

aim was to support a (possibly widespread)

community in constructing an ontology, as the

ontology represents a common view. He presents the

two web-based systems Tadzebao and WebOnto,

which complement each other. Tadzebao supports

asynchronous and synchronous discussions on

ontologies. WebOnto, on the contrary, provides

features for collaborative browsing, creation and

editing of ontologies.

The systems Ontoverse (Weller, 2006) and Onto-

Wiki (Hepp et al., 2006) follow a wiki-based

approach. This approach explicitly focuses on the

collaborative construction of ontologies and supports

the cooperative design process. Term meanings and

problems can be discussed through the system by the

user community before these terms are integrated

into the system.

An important difference between these two

systems is that Ontoverse has a role-based concept

and restricts the usage to specific user groups. It

allows parallel development of ontologies with

different user groups in separated workspaces. The

access to OntoWiki on the other hand is open and

not limited. OntoWiki is realised using a standard

Wiki, whereas Ontoverse developed a new wiki-

based platform supporting the different steps of an

ontology development process: a planning and

conceptualisation phase; a phase of editing the

structural data; and finally a maintaining phase,

where updates, corrections and ontology

enrichments are performed (Weller 2006). Both

systems can be used also, to further evolve

ontologies.

Holsapple and Joshi (2001) suggest applying a

structured procedure for the collaborative working

process. This procedure uses a Delphi-like method

which incorporates feedback loops within the group.

Step by step the terms of the ontology are enriched

and better understood. In contrast to the wiki-based

systems before, he does not suggest a support

system.

Computational methods analyse texts and aim at

detecting changes or trends in the terms. This field is

known as emerging trend detection or topic

evolution (Gohr et. al., 2009), (Kontostathis, 2004).

The new terms serve as candidates for the ontology

enrichment. An interesting approach is presented by

Kruse (2009), even though not directly related to

ontology evolution. He presents an interview based

decision support system to identify drifts in the

VISUAL TREND ANALYSIS - Ontology Evolution Support for the Trend Related Industry Sector

meaning and usage of terms. The interview results

span a semantic room of meanings, which is

represented in a 3D visualisation for each person.

Kruse performed an interview with 100 people in

2007 and performed the same study again in 2008.

He asked the people to state their attitude regarding

car brands, definition of status and moral concept as

well as mobility in their daily life. Comparing the

result of these two studies he identified a change in

the definition of status. Status, once directly related

to big, representative and expensive cars, is no

longer connected to these things. It is rather linked

with environmental protection, social justice and

appropriate functionality. A rather small car is

preferred by the people in their daily life. Already in

2007 Kruse identified these changes in the

preferences, long before the market realized the

change in the car branch. A year later, 2008, the

press finally stated that they have problems selling

big and luxury cars. This is the obvious result of the

preference changes, already identified in the surveys

by Prof. Kruse.

Other approaches go further and do not only

extract term candidates but try to discover

conceptual structures, also. Maedche and Staab

(2000) present a framework that combines different

methods to acquire concepts, and to identify

relationships between concepts from natural

language. Relationships both relevant and not

relevant for the ontological taxonomy can be

identified. They applied their system in the domain

of telecommunications. For mining a concept

taxonomy in the first step they used domain specific

dictionaries. Further relations between concepts are

discovered based on generalised association rules.

These approaches aim at an (at least partial)

automation of ontology construction and are

subsumed in the field of ontology learning.

In our approach we explicitly rely on the experts

and give support trough intuitive visualisation

methods. The experience of Prof. Kruse with his

system shows that changes are (unconsciously)

realised by people before the changes reach printed

media and are then realised consciously by the

market or the public. Hence, the experts are an

important knowledge source. The experts know their

domain perfectly well and may have gut feelings and

ideas how things are developing. The graphical

visualisation on the other hand, supports the expert

user to discover changes in domain specific

knowledge in printed media and to identify missing

concepts in the ontology on this way. Using the tool,

they get means to monitor trend developments more

efficiently by the visualisation of domain specific

knowledge evolution. They can test their ideas and

push them into the market to set new trends.

3 OUR APPROACH

IN THE HOME TEXTILE

SECTOR

3.1 Requirements

The home textile sector is a trend-related industry.

Producers have to flexibly adapt to changing

preferences and consumer behaviour. Companies

and especially market analysts have to monitor

particular fields for recent trends that may impact

the company sales and success. For them, it is

important and required to detect emerging topics

early, and the way they evolve over time. The earlier

the producers identify particular techniques as well

as new colour and material combinations for a new

design, the better they are able to refine the products

and attract the interest of the customer. The

producers are able to surprise the customer

positively by new designs. On the one hand the

companies have to seek for trends while on the other

hand having to set those by varying colouring,

colour families, materials and their combinations to

create novelty and charm (Sauerwein, 2009). These

products are mainly sold through innovation and

design, nowadays. (Becella, 2007).

To address these requirements and provide

flexible analysis of trend combinations according to

the working tasks of designers and marketing

experts, we developed an ontology-based system

which allows analysing trend-relevant data sources

using text mining technology. By accessing digitised

fashion and trend magazines we enable a systematic

evaluation of colour families and material groups

which are mentioned in these sources. Marketing

experts and designers can analyse the frequencies of

colours, colour families, or material types from

magazines and trend books and assess the

development of colour and material statistics over

time.

An initial domain ontology was created that

serves as a starting point to detect trends and

monitor the market relevant sources. This well

known knowledge about concepts like colour,

material, structure or design of surface is modelled

in the ontology and forms the stable “anchors” to

identify the terminological development, which is a-

priori unknown. Marketing experts and designers

recognise new terms, that evolve from the active use

of these terms and concept combinations, and

supervise their development over time. These terms

may describe colours or surface structures, dominant

in magazines or articles and supervise their

development over time. These detected emergent

terms currently not incorporated in the domain

ontology are concept candidates for the ontology.

ICEIS 2010 - 12th International Conference on Enterprise Information Systems

Integrated into the ontology, they can be used for a

more specific search on particular trends to further

ensure reliability, accuracy and effectiveness of the

search (Klein and Fensel, 2001). Step by step the

domain ontology evolves to a trend ontology. The

search and evolution cycle is shown in Figure 1.

Support the search

Recognise

topic trends

Term in context

Ontology

Figure 1: Ontology based search for trends.

A system which supports designers and

marketing experts in this sector should meet the

following requirements:

 Systematic evaluation of colour families and

material groups which are mentioned in fashion

and trend magazines

 Assessment of the development of colour and

material statistics over time

 Flexible access to trend information according to

work tasks of designers (focus on particular

areas, colours, material, etc. as well as focus on

specific magazines)

 Intuitive means and features to modify the

ontology for an iterative evolution

In (Becks, 2007) we reported on the overall

design of the Trend Analyser system in the

AsIsKnown project. The next sections describes the

system features of our Term-Context-Language

Analysis (TeCLA) component in more detail,

followed by an evaluation in closed cooperation with

marketing experts and designers of a carpet producer

(cf. section 4). It combines the strengths of

automatic text analysis for a visual, goal-oriented

overview on trends with domain experience for

iterative and efficient evolution of market specific

trend ontologies.

3.2 System Overview

TeCLA is a semi-automatic, visual component,

supporting the experts in monitoring trend relevant

concept in relevant fashion magazines. It allows

intuitive access to upcoming colour and material

combinations. After the identification of trends the

community of experts (marketing specialists,

designer and trend scouts) is able to evaluate the

identified drifts in the concepts, match it with known

concept model in the domain ontology and possibly

adapt it accordingly. The experts are thus able to

form their trend specific ontology step by step. It is

especially important to include the domain experts in

the process of ontology evolution for several reasons

(Fensel, 2001b), (Hepp et al. 2006). Firstly, the

ontology aims at a consensual domain understanding

and knowledge. Hence, the evolution process

requires cooperation and exchange of information

between different people, such as domain experts.

The strength of the community cannot be replaced

by a single ontology expert. Secondly, the process

stays under control of the community, which

enhances the acceptance of the resulting ontology.

The experts are the ones who are working with the

ontology. They can evolve ideas and form the

ontology using their high degree of experience and

implicit background knowledge. The process of

trend detection is always related to experience and

creativity. A fully automatic approach would not be

accepted by such a user group, as the process is not

under their control. Additionally, our interactive,

user integrated approach meets their rather creative

and weakly structured way of working.

3.2.1 Visualisation and Change Detection

The first step in using TeCLA is the configuration of

the analysis matrix for a particular trend detection

(cf. Figure 2, showing the analysis result for the

concepts “green” and “brick” in AIT). The analysis

matrix defines the group of magazines and articles

as well as the period and aggregation level of time

for which trend relevant concepts shall be analysed

in the magazines. The trend relevant concepts are

selected from the ontology by the user e.g. colours,

materials, architects. Based on these concepts

TeCLA computes the concept term co-occurrences

and represents them as term context stars (shown in

the matrix cells in Figure 2), one for each selected

concept in each cell of the analysis matrix.

The terms appearing in the context of the

selected concept are determined using methods of

natural language processing. Magazines are first

linguistically pre-processed: The tokenised texts are

automatically annotated with part-of-speech tags that

indicate the grammatical categories of each word.

By means of dictionaries, for each known term a

matching concept from the ontology is attached

(word sense tagging). See (Simov et al., 2004) for

more details on the web-service based architecture

used.

As soon as the user has provided the list of

concepts that shall be analysed in the magazines,

VISUAL TREND ANALYSIS - Ontology Evolution Support for the Trend Related Industry Sector

commit change

new concept the user specifies which concept of the

ontology should be the super concept. He types in a

name for the new concept (the URI is added from

the system in the background), as well as the

definition. In case of a change in the meaning of a

term the user selects the option change term.

Suppose the marketing and designers identified that

there is a shift in the colour of brown. The tone

formerly called brown is now called shilf. Therefore,

the expert selects the concept brown and types in the

new term (cf. Figure 4). After committing the

change by clicking the button, the change is directly

visible in the ontology. In case of adding a new

concept clicking the commit button means that a

change request is send to an ontology expert. The

ontology expert checks for the required update and

possibly commits the change. The new version of

the ontology is then made available to the system.

4 EVALUATION

To evaluate the success and usability of our system

we used observational methods such as thinking

aloud protocols and structured interviews to assess

the functionality, usefulness and usability of the

system. The users are marketing experts and

designers from a carpet producer. They were given a

few minutes system presentation to be able to use

the features according to the given tasks during the

evaluation.

In the testing scenario the users were asked to

perform a whole walk-through of the system. They

performed an analysis of different trend relevant

concepts to identify emerging and descending trends

and were asked to use the different features

provided, to support their work. They were asked to

write down their “trend findings”. Afterwards they

were given another questionnaire to evaluate

especially the usability and usefulness of the

different features and the system as a whole.

The tests have shown that term context stars are

a good visualisation to represent the term co-

occurrences. Trend lines (increasing and decreasing)

can be identified quite easily based on the colour and

size of a term bubble of a context star. Common

terms in different stars are easily recognised and

accessed by using the comparison functionality of

TeCLA. A general remark from the users was to

change the colour coding of term highlighting in the

different features. Instead of using an additional

colour to mark the relevant terms, the irrelevant

terms should be shaded in grey. As a result the

important terms can be identified easier.

The overall results given by the users were

positive; they would all use the tool in similar

contexts again. The provided interaction possibilities

and features as well as the usability were evaluated

as good. The training prepared them well in using

TeCLA for the given analysis tasks. Nevertheless,

some result presentations can be improved to

facilitate the access to the information.

5 CONCLUSIONS

AND OUTLOOK

In this paper we presented an explorative approach

for the stepwise evolution of a trend ontology in a

creative, trend-related domain such as the home

textile industry. The text mining analysis is coupled

with a high degree of user interaction and

community-oriented modification of the ontology. In

addition to the related systems for ontology

evolution as presented in section 2 our system

provides and focuses on an intuitive visualisation

that supports the trend observation over a period of

time. Popular concepts and co-occurring terms in the

texts can be recognised and analysed over time. The

identified drifts give the experts an overview on the

market changes and help to evolve the ontology. The

system provides means to modify the ontology

accordingly. Thereby the ontology can evolve with

the domain thus better supporting the designer in

searching the magazines efficiently for further trends

and trend changes.

The community in this case is formed by marketing

experts, designers, trend scouts and product

managers. This is a group of domain experts with

sound knowledge working closely together. They

evaluate recognised upcoming and descending terms

and decide what should be modified in the ontology

accordingly. We do not provide collaboration

support in the system, explicitly, as this group of

experts is small and usually closely working

together.

During the whole development cycle informal

feedback was collected from our project partners

from the domain of carpet production several times.

In these evaluation phases it turned out early that an

explorative, community based approach with a high

degree of user involvement is necessary to establish

trust in the analysis results. The structured

evaluation proved this approach again. The user

interaction helps to derive ideas on possible trend

lines, taking into account the experience and

background knowledge of the experts. In this

ICEIS 2010 - 12th International Conference on Enterprise Information Systems

practical approach, we bring together the two

aspects of a trend experience and intuition of experts

as well as observation and analysis of relevant

market sources.

As mentioned in the beginning the aspect of

ontology versioning is closely related to the topic of

evolution. In the trend sector, it is also helpful to

have access to older versions of the ontology and to

be able to follow the changes made in the ontology.

Hence our work will focus on a history feature in the

next step. This feature will list all the changes which

were applied to get the current ontology. All the

changes can then be retraced by the experts and

provide additional information on the evolution of

trend concepts.

ACKNOWLEDGEMENTS

The research presented in this paper has been funded

by the AsIsKnown project

(http://www.asisknown.org) within the Information

Society Technologies (IST) Priority of the 6th

Framework Program (FP6) of the EU, and by the

Research School within the Bonn-Aachen Int.

Centre for Inform. Techn. (B-IT).

REFERENCES

Becella, M., 2007. Produkte verkaufen sich nur über

Innovation und Design, BTH Heimtex

Becks, A., Huster, J. 2007. Trend Analysis Based On

Explorative Data and Text Mining – A DSS for the

European Home Textile Industry, Proc. of 9th Int.

Conf. on Ent .IS (ICEIS), Funchal, Madeira.

Domingue, J., 1998. Tadzebao And Webonto: Discussing,

Browsing, Editing Ontologies On The Web, 1th Know.

Acq. for Know.-Based Systems Workshop, Banff,

Canada

Fensel, D. 2001a. OIL: An Ontology Infrastructure for the

Semantic Web IEEE Intelligent Systems, 16(2)

Fensel, D. 2001b. Ontologies: Dynamic Networks of

Formally Represented Meaning Proc. of the 1st

Semantic web working symposium, California, USA

Flouris, G.S., Manakanata, D.S., Kondylaki, H.S.,

Plexousaki, H. S., Antonio, G.U., 2008. Ontology

change: classification and survey The Knowledge

Engineering Review 23(2)

Gohr, A., Hinneburg, A., Schult, R., Spiliopoulou, M.,

2009. Topic Evolution in a Stream of Documents,

Proc. of the SIAM Int. Conf. on Data, Nevada, USA

Haase, P., Sure, Y. 2004. State-of-the-Art on Ontology

Evolution, Deliverable 3.1.1.b, Institute AIFB,

University of Karlsruhe

Hassan-Montero, Y. Herrero-Solana, V. 2006. Improving

Tag-Clouds as Visual Information Retrieval Interfaces.

Int. Conf. on Multidisciplinary Information Sciences

and Technologies, InSciT2006, Mérida, Spain

Hepp, M., Bachlechner, D., Siorpaes, K., 2006. OntoWiki:

community-driven ontology engineering and ontology

usage based on Wikis, WikiSym '06: Proc. of 2006

international symposium on Wikis, Odensee, Denmark

Heringer, Hans Jürgen 1998. Das Höchste der Gefühle –

Empirische Studien zur distributiven Semantik. Verlag

Stauffenburg, Tübingen

Holsapple, C. W., Joshi, K. D., 2002. A collaborative

approach to ontology design Com. of the ACM, 45(2),

pp. 42-47

Klein, M.., Fensel, D. 2001. Ontology versioning for the

Semantic Web, Proc. of the 1st Int. Semantic Web

Working Symposium, California, USA

Kontostathis, A., Holzman, L.E.. Pottenger, W.M., 2004.

Use of Term Clusters for Emerging Trend Detection,

Technical Report

Kruse, P. 2009. Ein Kultobjekt wird abgewrackt, GDI

Impuls, Wissensmagazin für Wirtschaft, Gesellschaft,

Handel, Nr. 1

Maedche, A., Staab, S., 2000. Mining Ontologies from

Text, Proc. of the 12th Int. Conf., EKAW, 189-202,

Juan-les-Pins, France

Sauerwein, T. 2009. Hans Joachim Schilgen,

Geschäftsführer Verband Deutsche

Heimtextilindustrie, Interview über Branchensituation,

Wahre Renner der Textilbranche und Nachhaltigkeit,

Online: http://dhd-news.de/article/3958/branchensitua

tion-in-der-heimtextilindustrie-o-t-ne

Simov, K., Simov, A. Ganev, H., Ivanova, K., Grigorov, I.

2004. The CLaRK System: XML-based Corpora

Development System for Rapid Prototyping. Proc. of

LREC, Lisbon, Portugal, pp. 235-238

Stojanovic, L., Maedche A., Motik B., Stojanovic N.

2002. User-Driven Ontology Evolution Management,

Proc. of the 13th Eu. Conf. on KE and KM (EKAW

2002), Madrid, Spain

Weller, K., 2006. Kooperativer Ontologieaufbau, Proc. of

28. Online Tagung der DGI, Frankfurt a.M., Germany

Welty, C., Lehmann, F., Gruninger, G.,Uschold, M. 1999.

Ontology: Expert Systems All Over Again? Invited

panel at AAAI-99: Nat. Conf. on AI. Austin, Texas.

VISUAL TREND ANALYSIS - Ontology Evolution Support for the Trend Related Industry Sector