Semantic and Syntactic Matching of e-Catalogues

Using Vector Space Model

Ahmad Mehrbod, Aneesh Zutshi and António Grilo

Unidade de Investigação em Engenharia Mecânica e Industrial (UNIDEMI), FCT/UNL, Campus de Caparica, Portugal

Keywords: e-Catalogue, Semantic and Syntactic Integration, Vector Space Model.

Abstract: E-catalogues play a key role in e-procurement. Matching a product request from a buyer with products e-

catalogues, helps companies to find business partners in e-marketplaces. Plethora of e-catalogue standards

and enterprise specific formats is one of the major challenges in the matching process. In addition to this

syntactic heterogeneity, the same product concept can be expressed using different words that causes

semantic diversity in e-catalogues. In this paper, we propose a flexible approach to e-catalogue matching

using Vector Space Model. Each e-catalogue is interpreted syntactically in its schema and semantically in

an ontology that is developed based on its product classification system. Schemas and ontologies are added

to the VSM by adding the syntax of the structure and semantic of the ontology to the indexing and searching

mechanisms. The matching process uses the syntactic and semantic metadata for interpreting each e-

catalogue as much as the information is available for the system, but is not dependent on this information.

1 INTRODUCTION

E-catalogues are electronic representations of

information about the products and services of an

enterprise (Huang and Huang, 2005). They play a

critical role in e-procurement marketplaces. E-

catalogues can be used in both the tendering (pre-

award) and the purchasing (post-award) processes.

Companies use e-catalogues to exchange product

information with business partners. Suppliers use e-

catalogues to describe goods or services that they

offer for sale. Meanwhile buyers may use e-

catalogues to specify the items that they want to buy

(Ghimire et al., 2013a) (Ghimire et al., 2013b).

Matching a product request from a buyer with

product e-catalogues that have been provided by the

suppliers, helps companies to reduce the efforts

needed to find partners in e-marketplaces(Lee et al.,

2007).

This research is based on a research and

development project in an e-procurement provider

company and provides a framework for matching

various e-catalogues originating from suppliers and

buyers in the e-procurement platform. In an e-

procurement e-marketplace at least two scenarios for

finding similar e-catalogues are possible (Ghimire et

al., 2013b) (Grilo et al., 2013a). First a buyer who

makes a call for tender needs to select some

suppliers based on their e-catalogues in order to send

the invitation. The second scenario occurs when a

supplier searches to find opportunities in e-

marketplace. A supplier may upload a product e-

catalogue to the search interface in order to find

similar call for tenders. In the search scenario, a user

has an e-catalogue and seeks similar e-catalogues in

the platform.

2 RESEARCH PROBLEM

The large variety of e-catalogue formats which are

used by various companies is a major challenge in

the matching process. Since each business actor may

use a different structure, classification and

identification code for describing e-catalogues, it is

not easy to match a product with the e-catalogue

requested by another partner (Lee et al., 2007). This

heterogeneity makes it difficult and time-consuming

to integrate and query e-catalogues (Chen et al.,

2010a).

While, there are too many different standards for

e-catalogues and product classifications in use, often

companies do not follow standard formats and prefer

to have their individual structures (Chen et al.,

2010b). Hence we often encounter a plethora of

catalogue formats (Grilo and Jardim-Goncalves,

224

Mehrbod A., Zutshi A. and Grilo A..

Semantic and Syntactic Matching of e-Catalogues - Using Vector Space Model.

DOI: 10.5220/0005115302240229

In Proceedings of the 11th International Conference on e-Business (ICE-B-2014), pages 224-229

ISBN: 978-989-758-043-7

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

2013) ranging from unstructured text to well-

structured XML documents. This diversity results in

the syntactic heterogeneity of e-catalogues.

Syntactic diversity is only one side of the

heterogeneity problem of matching e-catalogues.

Other and yet more complicated side of this problem

is the semantic diversity of e-catalogues. The same

product concept can be expressed using different

keywords, classifications or expressions. As result,

different users may use different terms to express a

same product concept, which makes the matching

process get different result when facing a synonym

query (Chen et al., 2010b).

3 STATE OF THE ART

The heterogeneity of e-catalogues which come from

various sources (Grilo et al., 2013b) causes difficulty

in the matching process. Generally we encounter

with two aspects of heterogeneity in e-catalogues

which are semantic and syntactic diversity. Syntactic

heterogeneity is the result of different document

structures and catalogue formats. Semantic

heterogeneity refers to the different meanings of the

words in various contents(Lee et al., 2007).

In order to encounter with the integration

problem of e-catalogues, several approaches and

methods have been proposed:

3.1 Standardization

In order to avoid the semantic diversity,

classification systems such as CPV, UNSPSC and

eCl@ss try to standardize the terms used for

describing goods and services which are the subject

of procurement. Using a common classification

system for products and services enables reliable and

efficient exchanges of product data across

organizations. Additionally, e-catalogue standards

such as UBL, BMEcat and cXML recommend using

of these classification systems and furthermore

propose common data structures for unifying e-

catalogue schemas usually for exchanging purposes.

However, catalogue standards and classification

systems are not sufficient to meet all the

requirements of data exchange. Consequently, often

enterprises do not follow standard formats and prefer

to have their individual structures(Chen et al.,

2010b). Also, variety of standards makes it

impractical to reach the classification and schema

unification goal. These standards differ in addressed

markets, capabilities to represent product

information, market acceptance, and standardization

processes (Schmitz et al., 2005). This problem is

more visible in multi-source e-marketplaces

(Ghimire et al., 2013a) (Grilo and Jardim-

Goncalves, 2013). There are at least 25 standards

relating to e-catalogue and product classification

systems, and thousands of enterprise products

databases and e-commerce sites(Kim et al.,

2007)(Chen et al., 2010b)(Schmitz et al., 2005).

3.2 Uniform Model

One traditional approach to solve the integration

problem is to transform different formats into a

uniform catalogue model (Ghimire et al.,

2013b)(Ghimire et al., 2013a)(Chen et al., 2010a)

that serves as reference format. In order to achieve

this general model, these approaches formulate a

formal model to represent various catalogues or

select an existing standard as the general model.

Then mapping functions have been designed that can

handle the transformation of different formats into

the uniform model.

But within this heterogeneous set of known or

even unknown structures achieving a uniform

structure for e-catalogues is usually not practical.

Development of a uniform e-catalogue model

requires precise and detailed understanding of each

of the various formats of catalogues. However, there

is always a chance to encounter with a new format

which may cause difficulties in its interpretation.

Furthermore for transformation to a uniform model,

e-catalogues must be completely validated and in

conformance to the expected format with no

tolerance from format deviations.

3.3 Ontological Model

Syntactic interoperability alone does not handle all

the problems of integration. Even if we get

structured product information, it does not guarantee

that we can interpret the content precisely when e-

catalogues use different taxonomies. So, ensuring

semantic interoperability is inevitable in the

interpretation of product information(Kim et al.,

2007). Therefore, several efforts such as (Huang and

Huang, 2005), (Chen et al., 2010a) and (Lee et al.,

2006) have encountered the problem by providing a

universal ontological model for product data.

The purpose is to introduce generic attributes to

design a universal ontology repository in order to

facilitate e-catalogue sharing and interoperability

(Chen et al., 2010b).The model is then used as a

standard reference for e-catalogue transformation or

development.

SemanticandSyntacticMatchingofe-Catalogues-UsingVectorSpaceModel

225

These ontologies are representation of products

and services which include the definitions,

properties, and relationships of the concepts that are

fundamental to products and services(Lee et al.,

2006). Usually these ontologies are constructed

based on semantic concepts of either a product

classification system or a product database. Many

companies classify products according to generic or

industry specific product classification standards, or

by using proprietary category systems. Such

classification systems often contain thousands of

product classes that are updated over time. This

implies a large quantity of useful product category

information for e-commerce applications. Thus,

instead of building up product ontologies from

scratch, which is costly, tedious, error-prone, and

high-maintenance, it is generally easier to derive

them from existing product classifications (Stolz et

al., 2014) (Lee et al., 2006).

Therefore these models have all the benefits of

standard schemas and product classification systems

and additionally improve accuracy of integration

process using semantic relationships. But they also

have the drawbacks of the standardization approach.

It seems impossible to have a globally accepted

reference model to create e-catalogues. Furthermore,

transforming product specifications to such models

not only is crucial and relies on manual efforts of

domain experts, usually led to inadequate results.

3.4 Ontology Merging

Generally, this approach is similar to uniform model

approach but targets the semantic integration instead

of syntactic integration. In this approach instead of

developing a universal ontological model to create e-

catalogues or transform them to a common world,

each e-catalogue is interpreted in its own ontological

model. Developing various models for each type of

e-catalogues is easier and more practical than

creating a universal model for covering all kind of e-

catalogues.

Due to different e-catalogue ontologies being

generated from different data sources are

heterogeneous, the key of semantic integration of e-

catalogue in this way turns out to be the mapping

and integration of catalogue ontologies (Chen et al.,

2010a)(Chen et al., 2010b). Therefore different

ontologies are combined into a new ontology that

includes and reconciles all the information from the

source ontologies according to semantic relations.

For example (Kim et al., 2007) designed a product

information mediation architecture by proposing

ontology mapping algorithm regarding both

taxonomy and attributes of underlying ontologies.

4 METHODOLOGY

4.1 Critique to Current Approaches

Based on two aspects of heterogeneity, syntactic

integration and semantic integration of multi-source

electronic catalogues have been attended to make e-

catalogues interoperable. Both of them require

integration of international product classification

standards, enterprise product databases and product

e-catalogue standards(Kim et al., 2007)(Chen et al.,

2010b).

Some previous research, such as from (Ghimire

et al., 2013a) and (Ghimire et al., 2013b) considered

more syntactic integration, others such as (Kim et

al., 2007) were more focused on semantic

integration and some such as (Huang and Huang,

2005) studied both at the same time. But regardless

of the semantic or syntactic dimension of the

problem, the general solution in e-catalogue

integration is to define a global model and convert e-

catalogues to this uniform model. Therefore these

traditional solutions either for semantic integration

or syntactic integration are dependent on universal

formal models.

4.2 Proposed Approach

Since in the area of e-catalogue we often encounter a

plethora of formats, and developing a universal

model is crucial, this research proposes to apply a

more flexible approach to the e-catalogues matching

problem. Vector Space Model (VSM) which is the

base of many search techniques and document

similarity methods can be applied to both semantic

(Mukerjee et al., 2011) and syntactic (Manning et

al., 2008) aspects of search problem. Although VSM

has been used in various search problems, its

application in matching e-catalogues is recent

(Mehrbod et al. 2014). In the context of current

research, VSM will be used to measure the syntactic

and semantic similarity ratio of providers’ e-

catalogues with a buyer’s e-catalogue.

Instead of developing semantic or syntactic

models and combining them to universal models that

try to cover all possible cases, the idea is to interpret

each e-catalogue syntactically in its schema and

semantically in an ontology that is made based on its

product classification system. Schemas and

ontologies will be added to the matching process by

ICE-B2014-InternationalConferenceone-Business

226

adding the syntax of the structure and semantic of

the ontology to the indexing and searching

mechanisms of VSM. The matching process uses the

syntactic and semantic metadata for interpreting

each e-catalogue as much as the information is

available for the system. Otherwise, it uses the basic

mechanism of VSM for the unknown formats.

4.2.1 Vector Space Model

VSM uses occurrence of keywords or terms in

documents to produce a table of vectors. Having a

vector model of documents, mathematical vector

operations can be applied to determine the similarity

of a document with another one or with a search

query. The simplest example is to use the deviation

angle between vectors of frequent terms to calculate

the relevance between text documents. While VSM

is used to deal with flat textual data, it is being

extended since the last two decades to treat complex

structured and semi-structured data (Tekli et al.,

2009).

4.2.2 Syntactic VSM

In order to encounter the syntactic heterogeneity

using VSM, we considered three general cases for e-

catalogues (Mehrbod et al. 2014). First, unstructured

text such as PDF files which are common in online

commerce. Second, structured or semi-structured

documents which are unknown for the system such

as enterprise specific formats. Third, structured

standard documents which are known for the system

such as cXML and UBL e-catalogues.

We used a Natural Language Processing tool to

extract terms from e-catalogues.

Then, a vector was

made to represent occurrence of terms in each e-

catalogue. E-catalogues that are similar to a given

search query can be calculated by comparing

deviation of angle between the vector of each e-

catalogue and that of the query.

In order to associate the syntaxes in calculating

similarity, levels of attributes in structured e-

catalogues are also included in the term definition.

Any structured or semi-structured document can be

shown using a tree.

Since content is distributed at

different levels of the tree, location of a term in the

tree is effective on the value of the term(Tekli et al.,

2009) and should be considered in term extraction.

Generally, values of attributes are disregarded in

term extraction from structured documents. This

approach is useful for structure-only comparing of

documents(Tekli et al., 2009). But in the context of

product features, similarity measure is more

sensitive to the values which have been saved in the

e-catalogue structures. Therefore in matching

process of e-catalogues, values are crucial and are

even more important than structures. Consequently,

we used a structure-and-content tokenization process

(Manning et al., 2008) to define the terms.

Figure 1: A part of a structured e-catalogue.

As an example, Figure 1 shows a portion of a UBL e-

catalogue which is used by PEPPOL (Ghimire et al.,

2013b). Table 1 shows the terms that are added to the

vector of this e-catalogue for the value paper in the

attribute name. Note that having values attached to all the

terms helps search process to avoid matching documents

with the same structure but different products.

In order to match e-catalogues that don’t have

completely the same structure with a lower

similarity ratio, we divided the weight of a term by

twice the number of nodes between the value and the

attribute. For detail information please refer to

(Mehrbod et al. 2014).

Table 1: Terms for paper in Figure 1.

Value Terms Ratio

Paper Paper

Name/Paper

Item/Name/Paper

CatalogueLine/Item/Name/Paper

Item/Paper

CatalogueLine/Item/Paper

CatalogueLine/Name/Paper

CatalogueLine/Paper

Standard e-catalogues are source of diverse types

of information. It usually includes general document

data, product data and partners’ data. This extra

information can mislead product search process.

Furthermore various attributes of product data can

have different value in the matching process. For

example classification code of a product has more

value than a description in matching process. Hence

we used tables of coefficients for known formats to

adjust the impact of each attribute in matching

process.

SemanticandSyntacticMatchingofe-Catalogues-UsingVectorSpaceModel

227

Figure 2: Coefficients for the sample e-catalogue.

Figure 2 shows the coefficients for the sample e-

catalogue of Figure 1. These coefficients are values

between 0 and 1 which are multiplied to the weights

of terms. Undesired information such as partners’

data can be simply excluded from matching process

by putting 0 coefficients. Using this simple

mechanism a new known structure can be easily

added to the search system. Default values for all

coefficients are 1 which reduces the status of an e-

catalogue to an unknown structure for the matching

process.

4.2.3 Semantic VSM

The general approach in semantic VSM is to expand

terms using their synonyms in an ontology. As

mentioned in the ontological modeling approach,

many efforts have been done to build up product

ontologies from existing classifications. These

ontologies are rich sources of semantic information

for interpreting product data. Recently, (Stolz et al.,

2014) developed a generic, semi-automated method

and tool for deriving OWL ontologies from product

classification standards and proprietary category

systems. Such approaches can be used to enrich

product descriptions with information from existing

data sources.

Hence it is straightforward to have an ontology

for semantic presentation of each e-catalogue. In the

case of standard formats, the relevant ontologies

have been published by making ontologies from

their product classification systems. For enterprise

specific e-catalogues, ontologies can be provided by

their companies using such available tools.

Then we have several overlapping ontologies in

product data domain. Each one is a different

abstraction and representation of the same or similar

concepts. As mentioned many efforts have been

done to integrate this ontologies using universal

models that is crucial. In order to encounter the

semantic heterogeneity, we want to empower our

VSM approach (Mehrbod et al. 2014) using a

simple, automatic and applicable ontology alignment

process based on modeling ontologies in a vector

space (Eidoon et al., 2008).

(Eidoon et al., 2008) developed a VSM for

finding similar entities from one ontology to entities

of another ontology without integrating them to a

common model. In this method, vectors representing

ontology concepts and properties are matched

iteratively and their similarity degree is estimated.

By expanding e-catalogue matching process

using this ontology alignment approach, we aim to

find synonym or similar terms from one e-catalogue

to ones of another e-catalogue. In this way the term-

vector of each e-catalogue will be expanded by

terms of all semantic concepts that exist in the e-

catalogue. Each distinct concept and property

represents one term in the vector space.

Adding these terms to the vectors and expanding

similarity calculation with ontology alignment will

enable the matching process to find semantically

similar e-catalogues.

5 CONCLUSIONS

This research improves a G2B2B e-procurement

platform by providing a semantic and syntactic

matching mechanism for e-catalogues. Several

companies use the e-procurement marketplace to

purchase and sell products and services. In this

process, e-catalogues can have an effective impact

on finding the right business partners. Heterogeneity

of e-catalogue formats that are used by various

business players is a significant barrier to the

application of the e-catalogues in searching

procedures.

Several researches have been done to solve this

multi-dimensional problem. In this paper we

reviewed and categorized these approaches into four

different classes and discussed their pros and cons.

Most of these research works have some limitations

for solving the problem including difficulties in

implementation that make them difficult to

implement.

In a previous work we had proposed an

alternate solution to cope with this problem using

flexible and practical method that is used by several

search engines. We applied Vector Space Model to

solve syntactic heterogeneity problems that we

encounter in matching e-catalogues. In this paper we

expanded the previous work to solve both syntactic

and semantic aspects of the heterogeneity problem.

First a brief review of the previous work on syntactic

matching mechanism has been presented. The

matching process had used a combinations of values

ICE-B2014-InternationalConferenceone-Business

228

and attributes of structured documents to find the

syntactic correlation of e- catalogues. A simple table

of coefficients was proposed to specify the matching

process for standard formats. This mechanism

increases the search precision by removing unrelated

information from the matching process and boosting

weights of important attributes.

Then a practical approach to expand the

matching mechanism was proposed. The proposed

extension improves the search process with semantic

matching using a simple, expandable and applicable

ontology alignment approach. The aim is to use the

benefits of all available ontologies and schemas but

not to be dependent on them.

The e-catalogue matching approach is

implemented as a prototype in the procurement

platform. The results show the matching process is

capable to match diverse formats of catalogues from

various sources. The experimental results are not

presented in this paper as the prototype is still on the

final stages for full production deployment.

ACKNOWLEDGEMENTS

The research of this work has been partially funded

by project VortalSocialApps, co-ﬁnanced by

VORTAL and IAPMEI and the European Funds

QREN COMPETE.

REFERENCES

Chen, D., Li, X., Liang, Y., Zhang, J., 2010a. A semantic

query approach to personalized e-Catalogs service

system. J. Theor. Appl. Electron. Commer. Res. 5, 39–

54.

Chen, D., Li, X., Zhang, J., 2010b. User-oriented

intelligent service of e-catalog based on semantic web.

2010 2nd IEEE Int. Conf. Inf. Manag. Eng. 449–453.

Eidoon, Z., Yazdani, N., Oroumchian, F., 2008. Ontology

matching using vector space. Adv. Inf. Retr. 4956,

472–481.

Ghimire, S., Jardim-Goncalves, R., Grilo, A., 2013a.

Framework for catalogues matching in procurement e-

marketplaces. In: Information Systems and

Technologies (CISTI), 8th Iberian Conference on. pp.

1–6.

Ghimire, S., Jardim-Goncalves, R., Grilo, A., Beca, M.,

2013b. Framework for inter-operative e-Procurement

marketplace. In: Computer Supported Cooperative

Work in Design (CSCWD), 2013 IEEE 17th

International Conference on. pp. 459–464.

Grilo, A., Jardim-Goncalves, R., 2013. Cloud-

Marketplaces: Distributed e-procurement for the AEC

sector. Adv. Eng. Informatics 27, 160–172.

Grilo, A., Jardim-Goncalves, R., Ghimire, S., 2013a. E-

Procurement in the Era of Cloud Computing. In:

Proceedings of the 4th International Conference on

Information Systems Management and Evaluation

(Icime 2013). pp. 104–110.

Grilo, A., Jardim-Goncalves, R., Ghimire, S., 2013b.

Cloud-Marketplace: New paradigm for e-

marketplaces. In: Technology Management in the IT-

Driven Services (PICMET), 2013 Proceedings of

PICMET ’13: pp. 555–561.

Huang, J.Z., Huang, G., 2005. Ontology-based e-catalog

matching for integration of GDSN and EPCglobal

network. IEEE Int. Conf. E-bus. Eng. 212–215.

Kim, W., Choi, D.W., Park, S., 2007. Agent based

intelligent search framework for product information

using ontology mapping. J. Intell. Inf. Syst. 30, 227–

247.

Lee, J., Lee, T., Lee, S., Jeong, O., Lee, S., 2007. Massive

Catalog Index based Search for e-Catalog Matching.

9th IEEE Int. Conf. E-Commerce Technol. 4th IEEE

Int. Conf. Enterp. Comput. E-Commerce E-Services

(CEC-EEE 2007) 341–348.

Lee, T., Lee, I., Lee, S., Lee, S., Kim, D., Chun, J., Lee,

H., Shim, J., 2006. Building an operational product

ontology system. Electron. Commer. Res. Appl. 5, 16–

28.

Manning, C.D., Prabhakar, R., Schutze, H., 2008.

Introduction to Information Retrieval. Cambridge

University Press.

Mehrbod, A., Zutshi, A. & Grilo, A., 2014. A Vector

Space Model Approach for Searching and Matching

Product E-Catalogues. In J. Xu et al., eds. Proceedings

of the Eighth International Conference on

Management Science and Engineering Management.

Advances in Intelligent Systems and Computing.

Lisbon, Portugal: Springer Berlin Heidelberg.

Mukerjee, K., Porter, T., Gherman, S., 2011. Linear scale

semantic mining algorithms in microsoft SQL server’s

semantic platform. Proc. 17th ACM SIGKDD Int.

Conf. Knowl. Discov. data Min. - KDD ’11 213.

Schmitz, V., Leukel, J., Dorloff, F., 2005. Do e-catalog

standards support advanced processes in B2B e-

commerce ? Findings from the CEN / ISSS workshop

eCAT 00, 1–10.

Stolz, A., Rodriguez-Castro, B., Radinger, A., Hepp, M.,

2014. PCS2OWL: A Generic Approach for Deriving

Web Ontologies from Product Classification Systems.

Semant. Web Trends Challenges SE - 43, Lecture

Notes in Computer Science 8465, 644–658.

Tekli, J., Chbeir, R., Yetongnon, K., 2009. An overview

on XML similarity: Background, current trends and

future directions. Comput. Sci. Rev. 3, 151–173.

SemanticandSyntacticMatchingofe-Catalogues-UsingVectorSpaceModel

229