Towards a Semantic Matchmaking Algorithm for Capacity Exchange in

Manufacturing Supply Chains

Audun Vennesland, Johannes Cornelis de Man, Peter Halland Haro, Emrah Arica and Manuel Oliveira

SINTEF, Trondheim, Norway

Keywords:

Semantic Matching, Semantic Matchmaking, Manufacturing Supply Chains.

Abstract:

Within supply chains, companies have difﬁculties in ﬁnding suppliers outside their known supplier pool or

geographical areas. The EU project MANUSQUARE aims to deploy a marketplace to match supply and

demand of supply chain resources to facilitate accurate and efﬁcient matchmaking. To this end, a semantic

matching algorithm has been developed as one of the key enablers of such a marketplace. The algorithm

exploits formal descriptions of resources provided by an ontology developed in the project and will later be

extended to incorporate additional data from different endpoints. This paper describes the main components

of the semantic matching algorithm, which on the basis of the formally described supply chain resources

returns a ranked list of relevant suppliers given a customer query. The paper further describes a comparative

evaluation of a set of common semantic similarity techniques that was conducted in order to identify the most

appropriate technique for our purpose. The results from the evaluation show that all four techniques perform

pretty well and are able to distinguish relevant suppliers from irrelevant ones. The best performing technique

is the edge-based technique Wu-Palmer.

1 INTRODUCTION

Complex and simple supply chains are continuously

undergoing improvements in terms of efﬁciency and

effectiveness, while exploring their integration with

both new and existing supply chains in related ar-

eas. A key challenge within supply chain usage per-

tains to identifying and mapping the right supplier at

the right time for the right service. Limited knowl-

edge and trust concerns often make purchasers depen-

dent on suppliers that are within reach: Local part-

ners, well-known names within an industrial sector,

Yellow-Pages or the ﬁrst hits on a search engine. To

establish supplier relationships outside a known do-

main or geographic area is therefore hard, creating

limited value networks by utilizing suppliers that are

limited to a speciﬁc geographic area and industrial do-

main. Although established supply chains are contin-

uously optimized i.e. “the process of completing ful-

ﬁllment functions efﬁciently and effectively” (Sherer,

2005), supply chains are not including suppliers that

better match the required services to create better

value networks, i.e. “link customer demand directly

to their networked supply chains” (Sherer, 2005). A

company could therefore beneﬁt from matchmaking

to establish supplier relationships outside its supplier

pool or geographic area.

Accurate and efﬁcient matchmaking of supply and

demand of manufacturing resources, including phys-

ical assets as well as human know-how, can have a

major economic impact on utilizing available capacity

for the right product at the right time. A purchaser’s

job is made more efﬁcient, using less time, and more

effectively, ﬁnding the right supplier that delivers the

right service at the right quality, and so on. Some

of the more recent solutions have turned to semantic

technologies for formal semantic descriptions that can

be interpreted by machines to identify semantic sim-

ilarity between offer and demand (Ameri and Patil,

2012; J

arvenp

a et al., 2018).

The EU project MANUSQUARE

aims to de-

ploy a European Platform-enabled marketplace facil-

itating matchmaking of supply and demand of man-

ufacturing resources. To support this objective, the

project has developed the MANUSQUARE ontol-

ogy for a formal representation of manufacturing re-

sources. This ontology incorporates abstract con-

cepts as well as domain-speciﬁc concepts capturing

the knowledge of industrial sectors addressed in the

project, such as manufacturing technologies solutions

and textile and cosmetics production.

With the ultimate goal of optimising the match

https://www.manusquare.eu/

466

Vennesland, A., Cornelis de Man, J., Haro, P., Arica, E. and Oliveira, M.

Towards a Semantic Matchmaking Algorithm for Capacity Exchange in Manufacturing Supply Chains.

DOI: 10.5220/0008364404660472

In Proceedings of the 11th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2019), pages 466-472

ISBN: 978-989-758-382-7

between supply and demand of manufacturing re-

sources, the semantic matchmaking algorithm consid-

ers two types of input:

1. Manufacturing resources offered by suppliers.

These resources are represented as RDF triples in

a knowledge base. The MANUSQUARE ontol-

ogy enables a formal description of the resources.

2. Various data provided by different endpoints that

can enhance the accuracy of the matchmaking and

offer added-value functionality. This can include

data from historical transactions to evaluate sup-

pliers along different dimensions, e.g. historical

matches, reputation indicators, consumer satisfac-

tion, etc.

The focus of this paper is on the ﬁrst type of input

and the technical development described is a seman-

tic matching algorithm that enables matching of offer

and demand based their representation in a knowledge

base according to the MANUSQUARE ontology.

The main contributions from this work are:

• An approach for a semantic matching algorithm

that matches resource demand with available re-

sources offered by suppliers.

• A summary of results from a comparative eval-

uation of different semantic matching techniques

used in the algorithm.

• Ideas on how supplementary data from differ-

ent sources could further enhance the results and

contribute to innovation in manufacturing supply

chains.

2 RELATED WORK

An initial literature study related to semantic match-

making in manufacturing logistics gave limited re-

sults

, however, we identiﬁed the three following pa-

pers as relevant. Ameri and Patil (Ameri and Patil,

2012) suggested a multi-agent framework based on

formal semantics for connecting buyers and sellers of

manufacturing services. In order to overcome the lim-

itations of pure string equality-based similarity tech-

niques, mediating agents match offered services with

requested capabilities based on taxonomy-based sim-

ilarity and feature-based similarity. The taxonomy-

based similarity uses subsumption reasoning to de-

termine if the ontological concept(s) representing a

query can be addressed by ontological concept(s)

representing available services. The feature-based

The literature search was conducted in Elsevier Scopus and

Google Scholar

similarity reﬁnes the results from the taxonomy-

based similarity by utilising the logical constraints de-

scribed by the concepts (e.g. via property restrictions)

in order to more accurately rank the relevant services.

A weighted set similarity measure, the Tversky Mea-

sure (Tversky, 1977), was used to compute a similar-

ity score from which the offered services are ranked.

arvenp

a et al. (J

arvenp

a et al., 2018) devel-

oped a capability matchmaking procedure for match-

ing product requirements with resource capabilities

and possible combinations thereof. The matchmak-

ing relied on a combination of ontologies allowing to

formally express product requirements and resource

capabilities, and business rules expressing more de-

tailed parameters such as dimensions of a given re-

source. The matchmaking between product require-

ments and resource capabilities consists of two con-

secutive steps: (1) Matching of product requirements

and resource capabilities at concept level. Product

requirements are represented as individuals of con-

cepts in the general capability ontology, and so are the

available resources. Hence, there is a match between

product requirement and available resources as long

as they are members of the same ontology concept;

(2) Detailed matching of parameters. This step em-

ploys the speciﬁed business rules and checks if there

is a match between parameters speciﬁed in the prod-

uct requirement and the offered resource (e.g., if a

screw type used by a screwing machine resource com-

plies with the required screw-type deﬁned in the prod-

uct requirement).

Sch

onb

ock et al. (Sch

onb

ock et al., 2018) used

matchmaking in the context of volunteering, i.e., con-

necting tasks with volunteers willing and capable of

performing them. An ontology coupled with meta-

information enabling a more explicit deﬁnition of ex-

pertise or task preference was used as a basis for the

matchmaking, resulting in a ranked list of tasks or

volunteers whose proﬁles match as closely as possi-

ble. The ontology included core aspects such as com-

petencies, spatio-temporal constraints and social re-

lationships. Tasks and volunteers are represented as

instances in the ontology and similarity values and

meta-information are linked to properties in the on-

tology. The matchmaking score between a given task

and a given volunteer is calculated based on (1) ag-

gregating the similarity values associated with the re-

lationships (properties) between concepts these in-

stances are members of (explicit similarity), (2) the

taxonomic structure of these concepts (implicit sim-

ilarity) and (3) meta-information such as how much

volunteers like/dislike a task, their level of expertise

and how important the task is. The former is based

on a ﬁxed similarity value representing the strength

Towards a Semantic Matchmaking Algorithm for Capacity Exchange in Manufacturing Supply Chains

467

of the relationship between two facet concepts (e.g.

equivalentTo has a higher similarity value than relat-

edTo), whereas the latter is based on semantic similar-

ity techniques considering e.g. the taxonomic prox-

imity and depth of the concepts.

3 APPROACH

The semantic matching algorithm described in this

section returns a ranked list of suppliers whose of-

fered resources match a consumer query. Both the

supplier resources and the consumer query are for-

mally represented by a set of ontology concepts de-

ﬁned in the MANUSQUARE ontology. The ontology

consists of over 1000 classes and a single resource

record as well as a query can be represented in a mul-

titude of ways. Figure 1 illustrates how a supplier

resource can be formulated using these facets in the

MANUSQUARE ontology. Here, the supplier Hack-

ett Group, which is situated in Penedo, Portugal, of-

fers welding of steel. This supplier can perform this

process between 01.10.2019 and 01.12.2019 with an

available capacity of 59 working hours. The company

is certiﬁed according to the ISO9001 quality manage-

ment standard.

Through discussion with academics and industry

representatives within the project, the following facets

are considered the most relevant parameters for the

semantic matching:

• Process (P): which manufacturing process is

sought by the customer?

• Material (M): which material does the consumer

want to have processed?

• Machine (MA): which machine / equipment is re-

quired for performing the process?

• Certiﬁcations (C): which company and/or qual-

ity management certiﬁcations are required by the

consumer?

• Capacity (CP): how much capacity (in terms of

production hours) is required to perform the re-

quested process?

• Calendar Availability (CA): Can the supplier de-

liver by the required due-date?

The facets are included as parameters in the fol-

lowing equation which returns a similarity score be-

tween a consumer query (q) and each individual re-

source registered in the knowledge base (r):

Sim(q, r) =

∑

x∈{F}

s f (x) (1)

F represents the set of facets values, and sf rep-

resents a similarity function as described in the fol-

lowing. The similarity score returned represents the

average of the sub-scores computed for each facet,

and when run on a dataset, the result is a ranked

list of suppliers whose offered resources match the

facets expressed in the query. Note that a consumer

can search for a combination of different processes

and leave blank uncertain ﬁelds in the query. This is

considered when computing the aggregate similarity

score.

For the facets Capacity and Calendar Availability

the similarity between a query and available resources

can be computed using simple Boolean matching, i.e.

either the registered supplier resource meets the con-

straints expressed by the consumer query (score 1.0)

or it does not (0.0).

For the facet Certiﬁcations, similarity is computed

using the Jaccard set similarity measure (Jaccard,

1901). The Jaccard similarity is computed by ﬁnd-

ing the intersected set of certiﬁcations from a query

and a supplier and dividing this by the union of certi-

ﬁcations.

For the facets Process, Material, and Machine

other similarity measures are needed. In order to com-

pute a similarity score between the query and avail-

able resources along these facets, techniques able to

exploit the taxonomic position as well as the context

of a resource should be employed. Such techniques

are typically categorised into edge-based and infor-

mation content-based techniques (Jiang and Conrath,

1997). Edge-based techniques consider the path

distance and taxonomic position of concepts to be

matched, while information content-based methods

are based on the assumption that the more information

two concepts share, the more similar they are. Here,

the information shared by two concepts is derived

from the information content of the concept(s) that

subsume them in the taxonomy. Information content

is quantiﬁed as negative the log-likelihood of ﬁnding a

given concept in a taxonomy and basically states that

the more abstract a concept is, the less information

it holds (Resnik, 1995). In our approach, we apply

a variant of information content called intrinsic in-

formation content (Seco et al., 2004; Pirr

o and Talia,

2010) which contrasts the “conventional” information

content by not relying on usage statistics of concepts

in a corpus. The intrinsic information content of a

concept c is computed as follows:

IC(c) = 1 −

log(Sub(c) + 1

log(|C|)

(2)

where Sub(c) indicates the number of subclasses of

the concept c and |C| represents the total number of

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

468

ProcessChain

ProcessType

Supplier

Period

MaterialType

Certiﬁcation

hasSupplier

hasPeriod

hasProcess

hasInput

hasCertiﬁcation

HACKETT_GROUP

«Penedo»

«Portugal»«Hackett Group»

hasNation

hasCity

hasName

hasQuantity

RESOURCE_X

PERIOD_X

01.10.2019

01.12.2019

hasFrom

hasTo

STEEL

WELDING

ISO_9001_HACKETT

hasInput

hasProcess

hasCertiﬁcation

hasSupplier

hasPeriod

INDIVIDUAL

CLASS

«Literal»

Supplier Resource

Figure 1: Ontological representation of a manufacturing resource.

concepts in the ontology.

The example illustrated in Figure 2 further ex-

plains the approach. For a given query, a similarity

score is computed from a pairwise matching of the

facets represented by the query and the facets repre-

sented by each resource stored in the knowledge base.

Once computed, these individual facet scores are then

averaged into a semantic similarity score representing

the semantic match between the query and a resource.

The facets represented within the dotted border are

matched using semantic similarity techniques exploit-

ing the structure of the ontology.

To determine which semantic similarity technique

to employ in the semantic matching an evaluation of

four different candidate techniques was conducted.

This evaluation is described in the next section.

4 EVALUATION

The evaluation is performed as a comparative evalua-

tion by running the algorithm described in Section 3

Query Resource

Process: CNCMilling

Material: MartensiticStainlessSteel

Machine: VerticalMillingMachine

Certiﬁcations: ISO9000, ISO9001, AS9100

Oﬀered Capacity: 75 hours

From date: 15.01.2019

To date: 01.04.2019

Process: Milling

Material: AlloySteel

Machine: MillingMachine

Certiﬁcations: ISO9000, ISO9004

Requested Capacity: 125 hours

From date: 01.02.2019

To date: 15.04.2019

Semantic Similarity: 0.63

0.94

0.67

0.91

0.25

0.00

1.00

Figure 2: Example of a semantic matching process.

in four different conﬁgurations (one for each semantic

similarity technique) on a dataset consisting of sup-

plier resources. The dataset was created as follows:

1. We selected a subset of concepts from the Mate-

rial branch in the ontology (Ferrous Metals).

2. From this subset relevant combinations of mate-

rial, process and machine were established on the

basis of rules speciﬁed by domain experts. A

rule could for example deﬁne that for the mate-

rial CarbonSteel, the process LaserBeamCutting

and the machine LaserCuttingMachine is a valid

combination.

3. For the remaining facets (Certiﬁcates, Capacity

and Calendar Availability) as well as for the sup-

plier data (ID, name, location) we simulated the

content using randomly generated input.

Using the above approach we generated 900 resource

record instances that were imported into the ontology

and used as our dataset. Figure 3 shows an excerpt of

two resource records from the test data.

Each of the four algorithm conﬁgurations used

one of the following semantic similarity techniques:

• Wu-Palmer (Wu and Palmer, 1994). The Wu-

Palmer algorithm is an edge-based method that

calculates a similarity score by considering the

depth of the two concepts to be matched (c

and

), along with the depth of their least common

subsumer (lcs):

Sim

2 ∗ depth(lcs)

(depth(c

) + depth(c

))

(3)

• Resnik (Resnik, 1995). Resnik is an information

content-based method that deﬁnes the similarity

Towards a Semantic Matchmaking Algorithm for Capacity Exchange in Manufacturing Supply Chains

469

Gaylord-Bechtelar Rakszawa Poland 125 ISO9001, LEED CarbonSteel Shaping EDMMachine

Supplier City Country Capacity Certiﬁcation Material Process Machine

15.01.19

Available from

01.04.19

Available to

Walsh LLC Berlin Germany 100 AS9000, MIL AluminumAlloy CNCMilling MillingMachine 01.02.19 15.03.19

Figure 3: Format of test data.

between two concepts as the information content

of their least common subsumer:

Sim

res

= IC(lcs) (4)

• Lin (Lin et al., 1998). Lin extends Resnik by in-

cluding a calculation of the information content of

the two concepts to be matched in addition to the

information content of their least common sub-

sumer:

Sim

lin

2 ∗ IC(lcs)

IC(c

) + IC(c

)

(5)

• Jiang-Conrath (Jiang and Conrath, 1997) propose

a hybrid approach that is derived from the edge-

based notion by adding the information content as

a decision factor. The normalised Jiang-Conrath

similarity (Seco et al., 2004) is computed as:

Sim

= 1 −

IC(c

) + IC(c

) − 2 ∗ IC(lcs)

(6)

Apart from using different similarity techniques, the

four conﬁgurations used the same approach, allow-

ing to isolate the performance measurement to the

similarity technique applied. The evaluation was per-

formed on a machine with Intel Core i7 processor and

16 GB of RAM memory. We generated a composite

consumer query that included two sub-queries rep-

resenting different and randomised variations of the

facets. Sub-queries reﬂect the fact that a consumer

may want to request multiple processes in one single

query, for example both cutting and assembling metal

parts.

For each of the four conﬁgurations, the 10 top re-

sulting hits returned by the algorithm were evaluated

for correctness by three domain experts. A majority

vote was used to consolidate the evaluation results,

hence if two out of three evaluators judged a result as

correct, it was ﬁnally considered correct. The eval-

uation measure used was precision@k (Elbedweihy

et al., 2015), whereby the precision is measured rel-

ative to the rank k of the search result. For example,

precision@3 is 0.67 if 2 out of the three ﬁrst search

results in the ranked list of search results are correct.

Since the experts only evaluated the top 10 search

results there is no full ground truth alignment from

which recall can be measured.

To support the domain experts in their evaluation

they were given some extra context information in the

form of a hierarchical listing of sub- and superclasses

for each of the ontology concepts relevant for each

query result. Since the domain experts had little ex-

perience with ontologies, such context information is

important for the validity and reliability of the evalu-

ation (Cheatham and Hitzler, 2014).

Figure 4 shows the results from the evaluation.

As the ﬁgure reveals, all four techniques returned

only correct search results among the top 3 suppliers.

Lin, Resnik and Wu-Palmer also achieves a 100 %

precision until the precision@6 threshold where Lin

and Resnik return 1 false positive result each while

Wu-Palmer maintains its 100 % precision. At preci-

sion@10, Wu-Palmer, the edge-based technique, ob-

tains the highest precision of 0.80. These results con-

tradict results from other experiments, e.g., those re-

ported in Resnik (Resnik, 1995) and Seco et al. (Seco

et al., 2004), where the information content-based

similarity methods perform better than edge-based

methods. One possible explanation of this is that

many of these experiments base the similarity compu-

tation on the WordNet ontology, which describes gen-

eral knowledge. When domain-speciﬁc ontologies are

used, as in our case, the results may differ (Pirr

2009).

There are some validity threats related to the eval-

uation that should be mentioned. First of all, and

in general, the task of assessing the relevance of

search results in information retrieval evaluation can

be highly subjective (Manning et al., 2010). In this

case the queries consisted of multiple parameters and

the threshold for determining their collective correct-

ness may vary. For example, should the domain ex-

perts weigh some parameters higher than others, or

should a search result be deemed correct if 5 out of 7

parameters are considered similar? Furthermore, the

experience and knowledge of the domain experts with

regard to particular details may also vary. For exam-

ple, one expert may be aware that a particular machine

is applicable to several different materials, while the

other experts may not.

Second, the domain experts used different strate-

gies for determining correct versus false search re-

sults. The ﬁrst domain expert required that both sub-

queries were fulﬁlled by a supplier’s offered resources

in order to state a correct result. The second domain

expert considered a search result as correct as long as

the resources offered by a supplier satisﬁed one out

of the two sub-queries. As did the third domain ex-

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

470

P@1 P@2 P@3 P@4 P@5 P@6 P@7 P@8 P@9 P@10

Jiang-C onrath 1,00 1,00 1,00 0,75 0,80 0,83 0,86 0,75 0,78 0,70

Lin 1,00 1,00 1,00 1,00 1,00 0,83 0,71 0,75 0,67 0,70

Re snik 1,00 1,00 1,00 1,00 1,00 0,83 0,86 0,88 0,78 0,70

Wu-Palmer 1,00 1,00 1,00 1,00 1,00 1,00 0,86 0,75 0,78 0,80

0,00

0,10

0,20

0,30

0,40

0,50

0,60

0,70

0,80

0,90

1,00

Precision

Evaluation results semantic similarity techniques

Figure 4: Evaluation Scores.

pert, however he used a strategy whereby the score

was graded into 0 (not correct), 1 (correct, but only

for one of the sub-queries) and 2 (correct for both

sub-queries). The four returned ranked lists of search

results contained mostly the same list of suppliers,

although ranked differently. In total there where 13

suppliers returned in the four lists of search results re-

turned (out of a total of 575 suppliers included in the

dataset). Looking at the consensus of the three do-

mains experts, all three were in perfect agreement on

5 of 13 the suppliers, 2 of the 3 experts agreed that

4 of the 13 suppliers offered relevant resources, while

the remaining 4 suppliers were considered relevant by

only 1 of the domain experts.

5 CONCLUSIONS AND FUTURE

WORK

5.1 Conclusions

This paper has described the development of a seman-

tic matching algorithm that will support matchmak-

ing between supply and demand of manufacturing re-

sources. The algorithm computes a similarity score

based on similarity along six facets which are repre-

sented both in a customer query and manufacturing

resources offered by suppliers. These facets are pro-

cess, material, machine, certiﬁcations, capacity and

calendar availability.

The algorithm employs formal descriptions

of manufacturing resources provided by the

MANUSQUARE ontology for the ﬁrst three

facets. As a step in selecting a technique that can

exploit these descriptions, we conducted a com-

parative evaluation of the four common semantic

similarity techniques Wu-Palmer, Resnik, Lin, and

Jiang-Conrath. The evaluation was performed by

three domain experts, who assessed four ranked lists

of search results returned from the algorithm using

each of the four techniques.

The results from the evaluation showed that Wu-

Palmer, an edge-based technique, obtained the highest

precision overall.

5.2 Future Work

As future work, the algorithm will be extended to also

consider other types of data from various endpoints.

These data will enable matchmaking by combining

the semantic matching presented in this paper with

direct and inferred trend analyses (e.g., from histor-

ical transactions), reputation benchmarking based on

user feedback, and analyses of collaboration patterns,

to name a few.

Although this paper focused on the terminologi-

cal part (the TBox) of the ontology, we also want to

exploit ABox capabilities once more instance data is

added. One idea along these lines is to use instance

matching as a means to correct erroneous data.

While the algorithm currently uses an unweighted

Towards a Semantic Matchmaking Algorithm for Capacity Exchange in Manufacturing Supply Chains

471

approach to compute similarity scores, we will inves-

tigate relevant weight conﬁgurations for the different

facets. This should be done in collaboration with rep-

resentatives from the supply chain industry to ensure

that the weights respond to the search strategies used

by those with a demand for supply chain resources.

By-products resulting from production could be

used in other production processes, but are often re-

garded as waste. The categorization of by-products

across different industrial sectors can lead to new

matches that were not thought of before. As part of

the future work, we will consider a categorization of

by-products for use in semantic matching.

Finally, a forthcoming and more comprehensive

evaluation will be conducted. Such an evaluation will

include a larger panel of domain experts to assess the

results and more concrete evaluation guidelines to re-

duce the possibility of validity threats promoted by

a clearer distinction between correct and false search

results.

ACKNOWLEDGEMENTS

The work presented here was part of the project

“MANU-SQUARE - MANUfacturing ecoSystem of

QUAliﬁed Resources Exchange” and received fund-

ing from the European Union’s Horizon 2020 re-

search and innovation programme under grant agree-

ments No 761145.

REFERENCES

Ameri, F. and Patil, L. (2012). Digital manufacturing mar-

ket: a semantic web-based framework for agile supply

chain deployment. Journal of Intelligent Manufactur-

ing, 23(5):1817–1832.

Cheatham, M. and Hitzler, P. (2014). Conference v2. 0: An

uncertain version of the oaei conference benchmark.

In International Semantic Web Conference, pages 33–

48. Springer.

Elbedweihy, K. M., Wrigley, S. N., Clough, P., and

Ciravegna, F. (2015). An overview of semantic search

evaluation initiatives. Journal of Web Semantics,

30:82–105.

Jaccard, P. (1901). Distribution de la ﬂore alpine dans le

bassin des dranses et dans quelques r

egions voisines.

Bull Soc Vaudoise Sci Nat, 37:241–272.

arvenp

a, E., Siltala, N., Hylli, O., and Lanz, M. (2018).

Product model ontology and its use in capability-

based matchmaking. Procedia CIRP, 72(1):1094–

1099.

Jiang, J. J. and Conrath, D. W. (1997). Semantic similarity

based on corpus statistics and lexical taxonomy. In

Proceedings of the 10th Research on Computational

Linguistics International Conference, pages 19–33.

Lin, D. et al. (1998). An information-theoretic deﬁnition of

similarity. In Icml, volume 98, pages 296–304. Cite-

seer.

Manning, C., Raghavan, P., and Sch

utze, H. (2010). In-

troduction to information retrieval. Natural Language

Engineering, 16(1):100–103.

Pirr

o, G. (2009). A semantic similarity metric combining

features and intrinsic information content. Data &

Knowledge Engineering, 68(11):1289–1308.

Pirr

o, G. and Talia, D. (2010). Ufome: An ontology

mapping system with strategy prediction capabilities.

Data & Knowledge Engineering, 69(5):444–471.

Resnik, P. (1995). Using information content to evaluate se-

mantic similarity in a taxonomy. arXiv preprint cmp-

lg/9511007.

Sch

onb

ock, J., Altmann, J., Kapsammer, E., Kimmerstor-

fer, E., Pr

oll, B., Retschitzegger, W., and Schwinger,

W. (2018). A semantic matchmaking framework for

volunteering marketplaces. In World Conference on

Information Systems and Technologies, pages 701–

711. Springer.

Seco, N., Veale, T., and Hayes, J. (2004). An intrinsic infor-

mation content metric for semantic similarity in word-

net. In Ecai, volume 16, page 1089.

Sherer, S. (2005). From supply-chain management to value

network advocacy: Implications for e-supply chains.

Supply Chain Management: An International Journal,

10:77–83.

Tversky, A. (1977). Features of similarity. Psychological

review, 84(4):327.

Wu, Z. and Palmer, M. (1994). Verbs semantics and lexical

selection. In Proceedings of the 32nd annual meeting

on Association for Computational Linguistics, pages

133–138. Association for Computational Linguistics.

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

472