Ontology Matching in Context-driven Collaborative Recommending

Systems

Alexander Smirnov, Alexey Kashevnik and Nikolay Shilov

St. Petersburg Institute for Informatics and Automation of the Russian Academy of Sciences,

39, 14 Line, 199178, St. Petersburg, Russia

Keywords: Context-driven Collaborative Recommending System, Ontology Matching, Ontology Alignment Pattern.

Abstract: The paper proposes an approach to building a context-driven collaborative recommending system, and

concentrates on the ontology matching algorithm and ontology alignment patterns. The designed

collaborative recommendation system is based on application of such technologies as user and group

profiling, context management, decision mining. It enables for self-organisation of user groups in

accordance with changing user profiles and the current situation context. Utilizing of the developed

ontology alignment patterns considerably accelerates the ontology fusion and matching processes due to

typification of fusion and alignment schemes.

1 INTRODUCTION

Integration of different Information Technologies

(IT) systems in flexible supply networks (FSN)

requires semantic integration of their data and

workflow models. This problem is even more

evident if IT systems of different enterprises (FSN

members) are considered. Developing frameworks,

with appurtenant models, needs to be based on solid

foundations. The alignment problem requires a

common ontology capturing business as well as IT

(Lind and Seigerroth, 2010).

Ontologies have shown their usability for this

type of tasks (e.g., (Bradfield et al., 2007); (Chan

and Yu, 2007); (Patil et al., 2005)). Integration of

different enterprise aspects into an ontology has

been also researched in a number of works. For

example, socio-instrumental pragmatism (Goldkuhl

and Röstlinger, 2002) incorporates human,

organizational, and IT enabled actions in a coherent

ontology. The concern of theorizing actions has also

been acknowledged by actor-network theory

(Latour, 1991), where technology and people are

both seen as social actants.

The service-oriented architecture (SOA) is a step

towards information-driven collaboration. This term

today is closely related to other terms such as

ubiquitous computing, pervasive computing, smart

space and similar, which significantly overlap each

other (Balandin et al., 2009).

The main idea of the approach is to represent

FSN elements by sets of services provided by them.

This makes it possible to replace the configuration

of FSN with that of distributed services. For the

purpose of interoperability the services are

represented by Web-services using the common

notation described by a common application

ontology (AO). Depending on the problem

considered the relevant part of AO is selected

forming the abstract context that, in turn, is filled

with values from the sources resulting in the

operational context. The operational context

represents the constraint satisfaction problem that is

used during self-configuration of services for

problem solving.

An intensive collaboration requires strong IT-

based support of decision making so that the

preferences from multiple simultaneous users could

be taken into account satisfying both the individual

and the group (McCarthy et al., 2006). Collaborative

recommending systems are aimed to solve this

problem. Recommending / recommendation

/recommender systems are widely used in the

Internet for suggesting products, activities, etc. for a

single user considering his/her interests and tastes

(Garcia et al., 2009), in various business applications

(e.g., (Hornung et al., 2009); (Zhena et al., 2009)) as

well as in product development (e.g., (Moon et al.,

2009); (Chen et al., 2010)). Collaborative

recommending is complicated by the necessity to

139

Smirnov A., Kashevnik A. and Shilov N..

Ontology Matching in Context-driven Collaborative Recommending Systems.

DOI: 10.5220/0004126801390144

In Proceedings of the International Conference on Knowledge Management and Information Sharing (KMIS-2012), pages 139-144

ISBN: 978-989-8565-31-0

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

Figure 1: Collaborative recommendation system architecture.

take into account not only personal interests but to

compromise between the group interests and

interests of the individuals of this group. In literature

(e.g., (Baatarjav et al., 2009); (Middleton et al.,

2003)) the architecture of the collaborative

recommending system is proposed based on three

components: (i) profile feature extraction from

individual profiles, (ii) classification engine for user

clustering based on their preferences (e.g.,

Romesburg, 2004)), and (iii) final recommendation

based on the generated groups. Development of

clustering algorithms capable to continuously

improve group structure based on incoming

information enables for self-organisation of user

groups (Flake et al., 2002).

2 CONTEXT-DRIVEN

COLLABORATIVE

RECOMMENDING SYSTEM

The developed context-driven collaborative

recommendation system architecture is presented in

Figure 1. It is centralized around the user clustering

algorithm (Smirnov et al., 2005) originating from the

decision mining area area (Smirnov et al., 2008);

(Rozinat and van der Aalst, 2006); (Petrusel and

Mican, 2010). The proposed clustering algorithm is

based on the information from user profiles. The

user profiles contain information about users

including their preferences, interests and activity

history. A detailed description of the profile can be

found in (Smirnov et al., 2009). Besides, in order for

the clustering algorithm to be more precise, this

information is supplied in the context of the current

situation (including current user task, product(s)

she/he works with, time pressure and other

parameters. The semantic interoperability (consistent

notation and terminology) between the profile and

the context is supported by the common ontology.

The user profiles are considered to be dynamic

and, hence, the updated information is supplied to

the algorithm from time to time. As a result the

algorithm can run as updated information is received

and update user groups. Thus, the development of

the algorithm has made it possible to be used for

building self-organisation mechanism for user group

formation.

Since, in the company considered as a case study

(Smirnov et al., 2011), the major difference between

users is the group of products they work with, the

generated groups are expected to be product related.

However, in other environments this is not necessary

to be the case and groups can be process-oriented,

resource-oriented or other.

When groups are generated the common

preferences/interests of the groups can be identified

based on the results of the clustering algorithm.

These preferences can be then generalized and

analyzed in order to produce collaborative

recommendations.

3 ONTOLOGY MATCHING

In order to analyze the existing ontology matching

techniques an extensive state-of-the-art review has

been done, which covered about 20

systems/approaches/projects related to ontology

matching (Smirnov et al., 2010).

All the similarity metrics in the performed state-

of-the-art review are based on the two information

retrieval metrics of precision and recall. The balance

between these is achieved via choosing the right

threshold value. The possibility of choosing the right

threshold value has to be taken into account in the

development of the matching models.

Since in enterprise information systems most of

services are problem-oriented, the usage of reusable

ontology patterns for the common ontology creation

is proposed. This would enable unification and

standardization of the ontologies and significantly

simplify the ontology matching.

Common ontology

Self-organization

Current

situation

described by

context model

Clustering

algorithm

User groups

based on

common

preferences /

interests

Identification of

common

preferences /

interests

Group

recommendation

User profiles

KMIS2012-InternationalConferenceonKnowledgeManagementandInformationSharing

140

Based on it the following concluding remarks

can be made.

The goal of ontology matching is basically

solving the two major problems, namely:

(i) ontology entities with the same name can have

different meaning; (ii) ontology entities with

different names can have the same meaning.

For this purpose a number of techniques are

applied in different combinations. These techniques

include:

 Identification of synonyms

 Similarity metrics (name similarity, linguistic

similarity)

 Heuristics (for example two nodes are likely to

match if nodes in their neighborhood also match)

 Compare sets of instances of classes instead

compare classes

 Rules: for example, if class A1 related to class

B1 (relation R1), A2 related to class B2 (relation

R2) and B1 similar to B2, R1 similar to R2 therefore

A1 similar to A2.

As a result of matching the following types of

elements mapping proximity can be identified:

 One-to-one mapping between the elements

(Associate-Professor to Senior-Lecturer)

 Between different types of elements (the relation

AdvisedBy(Student, Professor) maps to the attribute

advisor of the concept Student)

 Complex type (Name maps to the concatenation

of First Name and Last Name)

All methods can be separated into the following four

groups:

 Linguistic Methods are focused on determining

similarity between entities based on linguistic

comparison of these entities (count of the same

symbols estimation, estimation of the longest similar

parts of words, etc.).

 Statistical Methods (Instance-Based) compare

instances of the ontology entities and based on this

estimation entities can be compared.

 Contextual Methods are aimed at calculation of

a similarity measure between entities based on their

contexts. For example if parents and children of the

ontology classes are the same consequently the

classes also the same.

 Combined Methods integrate specifics of two or

three of the above methods.

In the considered problem domain the differentiation

between instances is not an easy task. Because of

this reason the techniques and methods relying on

instances were not considered for further

development. Hence, the developed models

presented below integrate all of the above techniques

(except those dealing with instances) and propose a

set of combined methods having features of the

linguistic and contextual methods.

4 MULTI-MODEL APPROACH

FOR ONTOLOGY MATCHING

The below proposed approach allows matching of

ontologies for the interoperability purposes and is

based on the ontology matching model illustrated by

Figure 2. The approach takes into account that the

matching procedure has to be done “on-the-fly” and

remembering the fact that matched ontologies are

responsible for performing certain concrete and

well-described tasks, which means that they

generally should be small–to–medium size and

describe only limited domains. A detailed

description of the approach and experimentation

results can be found in (Smirnov et al., 2010).

Ontology is represented as RDF triples,

consisting of the following ontology elements:

subject, predicate, object. Degree of similarity

between two ontology elements is in the range [0, 1].

The approach consists of the following steps:

 Compare ontology elements taking into account

synonyms of both ontologies. The degree of

similarity between equal elements is set to 1

(maximum value of the degree of similarity).

Figure 2: Multi-model approach to on-the-fly ontology matching.

Comparison of elements of two ontologies

using similarity-based method

Comparison of elements of two ontologies

using semantic-based distances search method

Graph-based distance improvement

Comparison of elements of two ontologies using synonyms

Linguistic

Contextual

Combined

Method class

Matching model

OntologyMatchinginContext-drivenCollaborativeRecommendingSystems

141

 Compare all elements between two ontologies

and fill the matrix M using similarity-based model.

Matrix M is of size m to n, where m is the number of

elements in the first ontology and n is the number of

elements in the second ontology. Each element of

this matrix contains the degree of similarity between

the string terms of two ontology elements using the

fuzzy string comparison method. At this step

WordNet or Wiktionary can be used for searching

semantic distances (Smirnov et al., 2010).

 Compare all elements of two ontologies and fill

the matrix M’. Matrix M’ is of size m to n, where m

is the number of elements in the first ontology and n

is the number of elements in the second ontology.

Each element of this matrix represents the degree of

similarity between two ontology elements.

 Update values in matrix M, where each new

value of elements of M is the maximum value of (M,

M’)

 Improve distance values in the matrix M using

the graph-based distance improvement model

(Smirnov et al., 2010).

As a result the matrix M contains degrees of

similarity between ontology elements of two

knowledge processors. This allows determining

correspondences between elements by selecting

degrees of similarities which are below than the pre-

selected threshold value.

5 ONTOLOGY ALIGNMENT

PATTERNS

Ontology alignment is a set of correspondences

between two or more (in case of multiple matching)

ontologies obtained as a result of the ontology

matching process (Euzenat and Shvaiko, 2007). In

this section the complicated ontology alignment

situations (patterns), which may arise during setting

relationships between elements and the rules of their

processing, are presented. These patterns are valid

for both straight and reverse directions.

Patterns are a proven way to capture experts’

knowledge in fields where there are no simple “one

size fits all” answers (Enterprise Integration

Patterns, 2012), such as knowledge fusion or

ontology alignment. Each pattern poses a specific

design problem, discusses the considerations

surrounding the problem, and presents an elegant

solution that balances the various forces or drivers.

In most cases, the solution is not the first approach

that comes to mind, but one that has evolved through

actual use over time. As a result, each pattern

incorporates the experience base that senior

integration developers and architects have gained by

repeatedly building solutions and learning from their

mistakes. This implies that patterns are not invented,

but discovered and observed from actual practice in

the field (Enterprise Integration Patterns, 2012).

Notations:

Source – ontology mapped;

Destination – ontology mapped to;

– class;

– attribute;

– associative relationship;

– hierarchical relationships or “class-attribute”

relationship;

– correspondence relationship.

Class-to-class Alignment. A class a" from the

Source corresponds (maps) to a class a' from the

Destination; a subclass b" of the class a" does not

correspond to any class from the Destination. In this

case search “in depth” does not stop and if a

subclass c" of the class b" corresponds to a class c'

from the Destination, the class c' becomes a subclass

of the class a', and the class c" becomes a subclass

of the class a". Experts can make a decision about

including or not the new class b" into the common

ontology (Figure 3).

Figure 3: Class-to-class alignment.

Attribute-to-attributes Alignment. An attribute

attr" of the class a" from the Source corresponds to

several attributes (a set of attributes) ATTR' of the

class a' from the Destination. In this case all the

attributes from ATTR' and methods for values

conversion should be added into the common

ontology (Figure 4).

Figure 4: Attribute-to-attributes alignment.

Class-to-classes Alignment. A class a" from the

Source corresponds to several classes (a set of

Source Destination

ATTR'

attr"

Source Destination

KMIS2012-InternationalConferenceonKnowledgeManagementandInformationSharing

142

classes) A' from the Destination. In this situation all

the classes from A' and conditions of selection

among these classes are added into the common

ontology. Attributes and subclasses of the class a"

are mapped into attributes or subclasses of the

classes from A' (Figure 5).

Figure 5: Class-to-classes alignment.

Class-to-attribute Alignment. A class a" from the

Source corresponds to a class a' from the

Destination; a class b" associatively connected to the

class a" from the Source corresponds to an attribute

attr' of the class a' from the Destination. In this

situation all the attributes and subclasses of the class

b" are mapped to the attribute attr' with appropriate

conversion methods and conditions are also added

(Figure 6).

Figure 6: Class-to-attribute alignment

Subclass-to-attribute Alignment. A class a" from

the Source corresponds to a class a' from the

Destination; subclass b" of the class a" corresponds

to an attribute attr' of the class a'. In this situation all

the subclasses of the class b" are mapped to the

attribute attr' or possibly to other attributes of the

class a', with appropriates conversion methods and

conditions being also added (Figure 7).

6 CONCLUSIONS

The major idea of the proposed approach to building

a context-driven collaborative recommending

system is FSN representation via a set of services

provided by its elements. SOA makes it possible to

abstract from real services and model these via Web-

services. Taking into account the described SOA

advantages this enables a higher level of abstraction

and ontology-based interoperability. The ontological

model is used to solve the problem of heterogeneity.

Figure 7: Subclass-to-attribute alignment.

The designed collaborative recommendation

system is based on application of such technologies

as user and group profiling, context management,

decision mining. It enables for self-organisation of

user groups in accordance with changing user

profiles and the current situation context.

The paper concentrates on description of the

developed multi-model approach to on-the-fly

ontology matching and ontology alignment patterns.

Utilizing of the patterns considerably accelerates the

ontology fusion and matching processes due to

typification of fusion and alignment schemes.

The presented work is yet in an early stage of

development. Only some of the proposed ideas have

been partially verified. The next step will be devoted

to application of the presented ontology alignment

patterns to a real world problem and further analysis

of their completeness and usefulness.

ACKNOWLEDGEMENTS

The research presented is motivated by a joint

project between SPIIRAS and Nokia Research

Center. Some parts of the work have been sponsored

by grants # 12-07-00298-a, # 12-07-00302-a, # 11-

07-00368-a, and # 11-07-00045-a of the Russian

Foundation for Basic Research, project # 213 of the

research program “Intelligent information

technologies, mathematical modelling, system

analysis and automation” of the Russian Academy of

Sciences, and project 2.2 “Methodology

development for building group information and

recommendation systems” of the basic research

program “Intelligent information technologies,

system analysis and automation” of the

Nanotechnology and Information technology

Department of the Russian Academy of Sciences.

Source Destination

attr

Source Destination

attr

Source Destination

a" A'

OntologyMatchinginContext-drivenCollaborativeRecommendingSystems

143

REFERENCES

Baatarjav, E.-A., Phithakkitnukoon, S., and Dantu, R.,

(2009). Group Recommendation System for Facebook.

In OTM 2008: Proceedings of On the Move to

Meaningful Internet Systems Workshop, LNCS 5333

(pp. 211-219).

Balandin, S., Moltchanov, D., and Koucheryavy, Y.,

(Eds.) (2009). Smart Spaces and Next Generation

Wired/Wireless Networking, Springer, LNCS 5764.

Bradfield, D. J., Gao, J. X., and Soltan, H., (2007). A

Metaknowledge Approach to Facilitate Knowledge

Sharing in the Global Product Development Process.

Computer-Aided Design & Applications, 4(1-4), 519-

528.

Chan, E. C. K. and Yu, K. M., (2007). A framework of

ontology-enabled product knowledge management.

International Journal of Product Development, 4(3-4),

241-254.

Chen, Y.-J., Chen, Y.-M., and Wu, M.-S., (2010). An

expert recommendation system for product empirical

knowledge consultation. In ICCSIT2010: The 3rd

IEEE International Conference on Computer Science

and Information Technology, IEEE (pp. 23-27).

Enterprise Integration Patterns, (2012). Retrieved from:

http://www.eaipatterns.com/.

Euzenat J. and Shvaiko, P., (2007). Ontology matching,

Springer-Verlag.

Flake, G. W., Lawrence, S., Giles, C. L. and Coetzee, F.,

(2002). Self-Organization and identification of Web

Communities, IEEE Computer, 35(3) 66-71.

Garcia, I., Sebastia, L., Onaindia, E., and Guzman, C.,

(2009). A Group Recommender System for Tourist

Activities. In EC-Web 2009: Proceedings of E-

Commerce and Web Technologies, The 10th

International Conference (pp. 26-37) Springer, LNCS

5692.

Goldkuhl, G. and Röstlinger, A., (2002). Towards an

integral understanding of organisations and

information systems: Convergence of three theories. In

Proceedings of the 5th International Workshop on

Organisational Semiotics.

Hornung, T., Koschmider, A., and Oberweis, A., (2009).

A Recommender System for Business Process Models.

In WITS’09: Proceedings of the 17th Annual

Workshop on Information Technologies & Systems.

Retrieved from: http://ssrn.com/abstract=1328244.

Latour, B., (1991). Technology is society made durable.

In: Law J. (Ed.) A sociology of monsters: Essays on

power, technology and domination, (pp. 103-131).

Routledge & Kegan Paul.

Lind, M. and Seigerroth, U., (2010). Multi-Layered

Process Modeling for Business and IT Alignment. In

Proceedings of the 43rd Hawaii International

Conference on System Sciences (HICSS), (pp. 1-10).

McCarthy, K., Salamo, M., Coyole, L., McGinty, L.,

Smyth, B., and Nixon, P., (2006). Group

Recommender Systems: A Critiquing Based

Approach, In IUI '06: Proceedings of the 11

international conference on Intelligent user interfaces

(pp. 267-269).

Middleton, S. E., De Roure, D., and Shadbolt, N. R.,

(2003). Ontology-Based Recommender Systems. In

Staab, S. and Studer, R. (Eds.) Handbook on

Ontologies (pp. 477-498) Springer.

Moon, S. K., Simpson, T. W., and Kumara, S. R. T.,

(2009). An agent-based recommender system for

developing customized families of products. Journal

of Intelligent Manufacturing, 20(6), 649-659.

Patil, L., Dutta, D., and Sriram, R., (2005). Ontology-

based exchange of product data semantics. IEEE

Transactions on Automation Science and Engineering,

2(3), 213-225.

Petrusel, P. and Mican, D., (2010). Mining Decision

Activity Logs. In Abramowicz, W., Tolksdorf, R., and

Wecel. K. (Eds.) BIS2010: Proceedings of Business

Information Systems Workshops, (pp. 67–79) Springer,

LNBIP 57.

Romesburg, H. C., (2004). Cluster Analysis for

Researchers, Lulu Press, California.

Rozinat, A. and van der Aalst, W. M. P., (2006). Decision

Mining in Business Processes, BPM Center Report no.

BPM-06-10.

Smirnov, A., Kashevnik, A., Shilov, N., Balandin, S.,

Oliver, I., and Boldyrev, S., (2010) On-the-Fly

Ontology Matching in Smart Spaces: A Multi-Model

Approach. In Balandin, S., Dunaytsev, R.,

Koucheryavy, Y. (Eds.) Smart Spaces and Next

Generation Wired/Wireless Networking, (pp.72-83)

Springer, LNCS 6294.

Smirnov, A., Levashova, T., Kashevnik, A., and Shilov,

N. (2009) Profile-based self-organization for PLM:

approach and technological framework. In PLM 2009:

Proceedings of the 6th International Conferecne on

Product Lifecycle Management.

Smirnov, A., Pashkin, M., and Chilov, N., (2005).

Personalized Customer Service Management for

Networked Enterprises. In ICE 2005: Proceedings of

the 11

International Conference on Concurrent

Enterprising (pp. 295-302).

Smirnov, A., Pashkin, M., Levashova, T., Kashevnik, A.,

and Shilov, N., (2008). Context-Driven Decision

Mining, In Wang, J. (Ed.) Encyclopedia of Data

Warehousing and Mining. Vol. 1. (pp. 320–327).

Hershey, New York, Information Science Preference,

Second Edition.

Smirnov, A., Shilov, N., Kashevnik, A., Jung, T., Sinko,

M., and Oroszi, A., (2011). Ontology-Driven Product

Configuration: Industrial Use Case, In Proceedings of

International Conference on Knowledge Management

and Information Sharing (KMIS 2011) (pp. 38-47).

Zhena, L., Huangb, G. Q. and Jiang, Z., (2009).

Recommender system based on workflow, Decision

Support Systems, 48(1), 237-245.

KMIS2012-InternationalConferenceonKnowledgeManagementandInformationSharing

144