Abstract

In last few decades, researchers have been motivated to facilitate ontology in several fields. In medical science, ontology is used to describe the theory of medical vocabularies and the correlation shared among them, thus permitting the sharing of medical knowledge. Definition based on ontology of disease allows each class of disease to be classified in a formalized structure singularly and along with the discussion of ontological realism for the treatment and diagnosis of disease. This paper uses a hybrid approach to evaluate ontology for epilepsy disease. A multi-criteria decision making (MCDM) method is used to decide the best ontology provided by Bio-portal to select the best suitable characteristics for epilepsy disease.

Keywords: Ontology, ontology evaluation, AHP, epilepsy, MCDM

13.1 Introduction

Nowadays, the single resource most critical for top management in an organization to sustain competitive advantage is knowledge. An organization can have competitive edge over its competitors by building an excellent process to manage knowledge. Ontology is a static knowledge representation method that includes definitions of basic concepts in a domain and relations among them which are machine interpretable. The information retrieval quality is improved from a keyword-based retrieval to knowledge-based search. Users should have a way for assessing ontologies and deciding which one fits their requirements to best face a multitude of ontologies. The main benefit of developing an ontology is that it enables sharing common understanding of the information structure and efficient reuse of domain knowledge.

Ontologies are an elementary data structure for knowledge conceptualization, but many ontologies are built for conceptualizing the same body of knowledge. Ontology selection is thus an important issue that must be addressed to access these and decide which one fits the requirements of the user best. Evaluating characteristics of ontologies has also become necessary because of the increasing number of candidates for reuse in a domain and complexities of ontologies. Various approaches for ontology selection have been considered in literature [6, 7, 9, 11, 20, 23]. A multi-criteria decision-making approach to ontology selection deals with the problem of selecting the ontology best suited to the needs of the decision maker.

Multi-criteria decision making (MCDM) assigns the selection of the choice of a best alternative from several available options in a decision, subject to several vague or concrete criteria or attributes. It helps people make decisions according to their choices in cases where more than one conflicting criterion exists. Analytical Hierarchy Process (AHP), an MCDM process, is a structured decision-making tool for quantifying the weights of decision criteria. Experience of experts is utilized to estimate the relative magnitude of parameters through pairwise comparisons. This paper presents a multi-criteria decision making approach based method to evaluate ontologies and select the best ontology in the medical domain.

13.2 Background

Methods have been developed to evaluate and select the best ontology. [16] explored the gain utilization of Semantic Web technologies in the domain of recruitment and developed an ontology based recruitment process. [17] used ontology evaluation techniques for agent cooperation to measure the quality of ontology. A criteria selection framework was proposed by [10] for guiding the selection of suitable criteria for various levels of ontology evaluation. [5] presented a method for ontology evaluation based on the goal, question, and metric approach for empirical evaluation. [15] proposed a method to integrate ontologies to select and recommend adapted internship seekers. [22] presented metrics, approaches, and other similar aspects of ontology evaluation in a concise manner.

[12] presented a scalable data-driven framework for ontology evaluation, targeting Big Data scenarios and use cases. [21] presented the outstanding contribution to ontology evaluation by considering social and community related themes. [2] organized four-categorical schemes to evaluate ontology in the existing literature. 200 ontology samples are considered, which are taken from the National Center for Biomedical Ontology (NCBO) Bio Portal. [4] evaluated the U Ontology by ontology evaluation methodology to evaluate the quality of the developed ontology. [11] discussed already present evaluation metrics that support the ontology evaluation process for offering guidance useful to knowledge handling, representation, and conceptualization. [13] explored different metrics used for the evaluation of quality of ontology from different dimensions for ontology evaluation. An approach to evaluate quality of reused parts was proposed by [24]. The model represented evaluation information with semantic properties.

[1] presented OnToology, a web-based application to manage ontology engineering support activities. [3] presented the 5 ontologist’s results, revealing high system usability of OntoKeeper and use-cases. [8] evaluated a set of consistent and objective ontology structural metrics. Ontology repositories used for evaluation have been used as corpora. [14] developed a decision support system for manufacturing process selection based on ontology-enabled case-based reasoning. From the review, it was observed that MCDM techniques are used to provide an efficient way to evaluate and select ontologies. This paper develops a method for evaluating and selecting ontologies in the medical domain based on the characteristics of a disease.

13.3 The Developed OntoEva Method

Ontology selection is the method of identifying the ontologies or ontology models that best suit the requirements of a decision maker. The precondition of ontology selection lies in evaluating all considered ontologies on the basis of certain criteria. The same body of knowledge is conceptualized by different ontologies, so the method developed in this paper will enable a user to select the one that best suits the requirements of the decision maker.

The OntoEva method was developed to evaluate and select the ontologies using AHP. The use of AHP will enable a decision maker in evaluating the ontologies based one some prioritized criteria. The proposed method determines the most suitable ontology by considering a set of characteristics and evaluation criteria. Weights are generated for each evaluation criteria according to pairwise comparison of the criteria and the criterion with the highest weight is considered as the most important criteria. Computation of the weights is guided by the experience of decision makers. OntoEva first identifies the characteristics of ontologies for a domain and computes weights for each criterion. The class mapping is then observed for prioritized factors, which helps in determining the best ontology for the domain. The methodology steps are as follows:

Step 1: Identify the information need.

Step 1.1: Identify the domain and the information need.

Step 1.2: Identify ontology structures that satisfy the information need.

Step 2: Specify the selection criteria. These criteria are similar to ontology popularity, topic coverage, or ontology structure.

Step 3: Identify n characteristics of ontology for a domain.

Step 4: Categorize these characteristics in m clusters.

Step 5: Construct a pairwise comparison matrix for each of the m clusters using the 9-point scale [18] and calculate (Inw) importance weights for the characteristics (criteria). Table 13.1 shows the 9-point scale is change to AHP pairwise comparison table.

Step 6: For the prioritized characteristics in each cluster, perform class mappings.

Step 7: Based on class mapping, compute importance weights for the considered ontologies.

13.4 Ontology Selection for Epilepsy Disorder

Epilepsy is a neurological disorder that affects people of all ages. According to [19], of the total population approximately 2% suffers from epilepsy, which ranks it as the second most common neurologic disorder. A person with epilepsy experiences symptoms like sensations, irritability, headache, depression, ‘funny feeling’, abnormal behavior, confusion, and sometimes loss of consciousness. The main reason of the suffering is lack of awareness about the disease. Thus, it becomes necessary that a commoner is provided a method to efficiently diagnose the disease at early stages. The three ontology structures considered in this study for this goal are:

• Extended Syndromic Surveillance Ontology (ESSO): To facilitate the mining of free-text clinical documents, an open-source terminological ontology is designed in English. It consists of epilepsy syndromes, seizure types, and data elements associated with them.
• Epilepsy and Seizure Ontology (EPSO): To support epilepsy focused informatics, tools this ontology are developed for patient care and clinical research.
• Epilepsy Ontology (EPILONT): It is ontology about the epilepsy domain and epileptic seizure based on the diagnosis proposed by the ILAE (International League Against Epilepsy).

The fundamental properties considered for the selection process are as follows.

13.4.7 Organizational Fitness

Organizational fitness involves various factors deciding the ease of how ontology can be deployed within an organization. A hospital may decide that all used ontologies align to the Bio-portal. This will help the organization in reducing costs when integrating data from different sources to align the ontologies.

The three ontologies, EPSO, EPILONT, and ESSO, fulfill all the above mentioned properties and are suitable for selecting the best ontology for epilepsy disease.

After selecting the ontologies, the topic coverage of these ontologies was analyzed from the Bio-portal, as given in Tables 13.2, 13.3, and 13.4. The characteristics analyzed were the ontology metric classes, properties, individuals, maximum depth, and maximum number of children.

In order to evaluate ontologies for epilepsy disease, the characteristics of epilepsy are selected after having discussions with doctors. Considering their opinions, the selected criteria are clustered under m=4 groups, as shown in Table 13.5, and were decomposed into hierarchy as illustrated in Figure 13.1.

Classes	1357
Individuals	2
Properties	29
Maximum depth	17
Maximum number of children	146

Classes	138
Individuals	0
Properties	10
Maximum depth	4
Maximum number of children	28

Classes	2705
Individuals	0
Properties	166
Maximum depth	12
Maximum number of children	214

For each of the clusters, a pairwise comparison matrix is developed (Table 13.6). These matrices were consistent, as the CI and CR values were < 0.1.

The computed weight for each characteristic is given in Table 13.7.

From the computed weights it can be seen that the focal type (A3) is the highest rated characteristic, AED (B2) is the highest weighted treatment, Head Trauma (C1) is the highest rated cause of epilepsy, and Confused Memory (D2) is the most noticeable symptom due to the highest weight value.

Cluster	Description	Criteria
[A] Types	Epilepsy is diagnosed in people when they have two or more seizures.	A1 – Absence A2 – Atomic A3 – Focal A4 – Generalized A5 – Tonic
[B] Treatment	The goal of treatment in patients suffering from epileptic seizures is to achieve a seizure-free status without any side effects.	B1 – Ketonic Diet B2 – AED B3 – Physical Exercise B4 – Surgery B5 – Nerve Stimulation
[C] Causes	The causes can be complex and sometimes hard to identify.	C1 – Head Trauma C2 – Brain Condition C3 – Prenatal Injury C4 – Alcohol Consumption C5 – Genetics
[D] Symptoms	Symptoms differ from person to person and according to the type of seizure.	D1 – High Fever D2 – Confused Memory D3 – Fainting D4 – Narcolepsy D5 – Cataplexy D6 – Panic Attack D7 – Breathing Difficulty

Cluster: Type
	A1	A2	A3	A4	A5
A1	1.000	0.333	0.200	0.333	0.333
A2	3.000	1.000	0.250	0.333	0.500
A3	5.000	4.000	1.000	4.000	4.000
A4	3.000	3.000	0.250	1.000	0.500
A5	3.000	2.000	0.250	2.000	1.000
Cluster: Treatment
	B1	B2	B3	B4	B5
B1	1.000	0.250	3.000	0.500	0.500
B2	4.000	1.000	3.000	2.000	2.000
B3	0.333	0.333	1.000	0.500	0.333
B4	2.000	0.500	2.000	1.000	2.000
B5	2.000	0.500	3.000	0.500	1.000
Cluster: Causes
	C1	C2	C3	C4	C5
C1	1.000	2.000	2.000	3.000	5.000
C2	0.500	1.000	2.000	2.000	3.000
C3	0.500	0.500	2.000	2.000	3.000
C4	0.333	0.500	1.000	1.000	2.000
C5	0.200	0.333	0.500	0.500	1.000

Cluster: Symptoms
	D1	D2	D3	D4	D5	D6	D7
D1	1.000	0.333	0.500	2.000	3.000	4.000	2.000
D2	3.000	1.000	2.000	2.000	3.000	4.000	5.000
D3	2.000	0.500	1.000	3.000	2.000	4.000	3.000
D4	0.500	0.500	0.333	1.000	0.500	3.000	2.000
D5	0.333	0.333	0.500	2.000	1.000	3.000	2.000
D6	0.250	0.250	0.250	0.333	0.333	1.000	3.000
D7	0.500	0.200	0.333	0.500	0.500	0.333	1.000

After identifying the prioritized characteristics for each cluster, each of these characteristics was searched in each of the three ontologies.

For Cluster I, Type, Focal was present in the class mapping.

These diagrams illustrate the presence of characteristic class in each of the ontologies.

EPSO Ontology (7-mapping)

ESSO Ontology (16-mapping)

Cluster	Criteria	Inw
A	A1	0.058
	A2	0.103
	A3	0.488
	A4	0.162
	A5	0.191
B	B1	0.128
	B2	0.373
	B3	0.081
	B4	0.230
	B5	0.188
C	C1	0.382
	C2	0.246
	C3	0.185
	C4	0.118
	C5	0.068
D	D1	0.166
	D2	0.298
	D3	0.213
	D4	0.098
	D5	0.113
	D6	0.060
	D7	0.052

EPILONT Ontology (0-mapping)

According to above data, the focal type of epilepsy exists in all the three ontologies. The focal type of epilepsy is the most prominent type among all the epilepsy types. Most patients suffering from epilepsy suffer from this type of the disease. In EPSO type presents with 7-class mapping, ESSO presents a focal class with 16-class mapping, and EPILONT presents with 0-mappping. Considering the above data, pairwise comparison matrixes are generated and Ontology Weights (Ow) are computed using the Eigenvector approach.

	EPSO	EPILONT	ESSO
EPSO	1.000	5.000	0.333
EPILONT		1.000	0.143
ESSO			1.000

The results are consistent, as the values of CI and CR were less than 0.1. The same procedure is performed for all other selected criteria.

For Cluster II, Treatment, the class mapping of each of the considered Epilepsy Ontology was observed. The Treatment AED is present in class mapping.

EPILONT (0 – mapping)

EPSO (7-mapping)

ESSO (1-mapping)

In EPSO, the AED class is present as a clinical drug component and its subclass drug brand name with 7-class mapping, ESSO consist of Medication_ Name class with 1-class mapping, and EPILONT consists of 0-mappping.

	EPSO	EPILONT	ESSO
EPSO	1.000	5.000	3.000
EPILONT		1.000	0.500
ESSO			1.000

For Cluster III, Causes, Head Trauma is present in class mapping as follows:

EPILONT – not found

EPSO (18 – mapping)

EPSO (20 – mapping)

In EPSO Ontology, Traumatic Brain Injury class is present with 20-class mapping, ESSO consists of the same class name with 18-class mapping, and in EPILONT it is not present.

	EPSO	ESSO
EPSO	1.000	1.000
EESO		1.000

For Cluster IV Symptoms, Confused Memory is present in class mapping:

EPILONT (71-mapping)

EPSO (0-mapping)

ESSO – Not Found

In EPSO Ontology, class seizure consciousness state is present with 0-class mapping, ESSO does not have that class, and in EPILONT it consists of a Confusion class with 71-class mapping.

	EPSO	EPILONT
EPSO	1.000	0.250
EPILONT		1.000

13.5 Results

On the basis of above analysis, the computed weights are shown in Table 13.8.

Finally, the average weights of each of the 3 ontologies are computed:

It can be seen that EPSO is the suitable ontology for diagnosing epilepsy disease.

13.6 Comparison of Ontologies

When these ontologies are compared from the Bio-portal platform on the basis of ontology metrics like classes, properties, individuals, maximum depth, and maximum number of children, the most suited ontology is ESSO.

However, from the computed weights, as shown in Table 13.8, EPSO was considered as the best ontology for epilepsy disease. The result was accepted by neurologists, as the EPSO provides all the required information a patient needs and it contains all the classes that best defined the epilepsy disease.

13.7 Conclusion

This paper presents a hybrid method for ontology selection. The method uses AHP technique to compute weights for determining the best suited ontology for diagnosing a disease. The study demonstrates its use for diagnosing epilepsy. Three ontologies, EPSO, ESSO, and EPILONT are considered for the study. Focal, AED, Head Trauma, and Confused Memory were evaluated as the important for determining the best ontology among these three.

For future reference, a more improved ontology for epilepsy disease can be developed. Ontology Evaluation and Selection can be done using other decision making techniques.

References

1. 1. Alobaid, A., Garijo, D., Poveda-Villalón, M., Santana-Perez, I., Fernández-Izquierdo, A., & Corcho, O. (2019). Automating ontology engineering support activities with OnToology. Journal of Web Semantics, 57, 100472.
2. 2. Amith, M., He, Z., Bian, J., Lossio-Ventura, J. A., & Tao, C. (2018). Assessing the practice of biomedical ontology evaluation: Gaps and opportunities. Journal of biomedical informatics, 80, 1-13.
3. 3. Amith, M., Manion, F., Liang, C., Harris, M., Wang, D., He, Y., & Tao, C. (2019). Architecture and usability of OntoKeeper, an ontology evaluation tool. BMC medical informatics and decision making, 19(4), 152.
4. 4. Ashraf, J., Hussain, O. K., Hussain, F. K., & Chang, E. J. (2018). Representation Phase: Ontology Usage Ontology (U Ontology). In Measuring and Analysing the Use of Ontologies (pp. 171-203). Springer, Cham.
5. 5. Bandeira, J., Bittencourt, I. I., Espinheira, P., & Isotani, S. (2016). FOCA: A methodology for ontology evaluation. ArXiv preprint arXiv:1612.03353.
6. 6. Brank, J., Grobelnik, M., & Mladenic, D. (2005, October). A survey of ontology evaluation techniques. In Proceedings of the conference on data mining and data warehouses (SiKDD 2005) (pp. 166-170). Citeseer Ljubljana, Slovenia.
7. 7. Fonou-Dombeu, J. V. (2019, June). A Comparative Application of Multi-criteria Decision Making in Ontology Ranking. In International Conference on Business Information Systems (pp. 55-69). Springer, Cham.
8. 8. Franco, M., Vivo, J. M., Quesada-Martínez, M., Duque-Ramos, A., & Fernández-Breis, J. T. (2019). Evaluation of ontology structural metrics based on public repository data. Briefings in bioinformatics.
9. 9. Haghighi, P. D., Burstein, F., Zaslavsky, A., & Arbon, P. (2013). Development and evaluation of ontology for intelligent decision support in medical emergency management for mass gatherings. Decision Support Systems, 54(2), 1192-1204.
10. 10. Hooi, Y. K., Hassan, M. F., & Shariff, A. M. (2015, May). Ontology evaluation—A criteria selection framework. In 2015 International Symposium on Mathematical Sciences and Computing Research (iSMSC) (pp. 298-303). IEEE.
11. 11. Jarosław, W. (2018). An Attempt to Knowledge Conceptualization of Methods and Tools Supporting Ontology Evaluation Process. Procedia computer science, 126, 2238-2247.
12. 12. Knoell, D., Atzmueller, M., Rieder, C., & Scherer, K. P. (2017). A Scalable Framework for Data-Driven Ontology Evaluation. In WM (pp. 97-106).
13. 13. Lourdusamy, R., & John, A. (2018, January). A review on metrics for ontology evaluation. In 2018 2nd International Conference on Inventive Systems and Control (ICISC) (pp. 1415-1421). IEEE.
14. 14. Mabkhot, M. M., Al-Samhan, A. M., & Hidri, L. (2019). An Ontology-Enabled Case-Based Reasoning Decision Support System for Manufacturing Process Selection. Advances in Materials Science and Engineering, 2019.
15. 15. M’Baya, A., Laval, J., Moalla, N., Ouzrout, Y., & Bouras, A. (2016, December). Ontology based multi criteria recomanded system to guide internship assignment process. 3rd IEEE International Conference on Computational Science and Computational Intelligence (CSCI 2016), Dec 2016, Las Vegas, United States. ffhal-01788507f
16. 16. Mochol, M., Oldakowski, R., & Heese, R. (2004). Ontology based recruitment process. Informatik 2004, Informatik verbindet, Band 2, Beiträge der 34. Jahrestagung der Gesellschaft für Informatik eV (GI).
17. 17. Plinere, D., & Borisov, A. (2014). Evaluation of the Ontological Knowledge Model/Ontoloģiskā zināšanu modeļa novērtēšana/Оценка качества онтологической модели знаний. Information Technology and Management Science, 17(1), 81-85.
18. 18. Saaty, T. L. (1988). What is the analytic hierarchy process? In Mathematical models for decision support (pp. 109-121). Springer, Berlin, Heidelberg.
19. 19. Stafstrom, C. E. (1998). The pathophysiology of epileptic seizures: a primer for pediatricians. Pediatrics in review, 19(10), 342.
20. 20. Sun, L., Ma, J., Zhang, Y., Dong, H., & Hussain, F. K. (2016). Cloud-FuSeR: Fuzzy ontology and MCDM based cloud service selection. Future Generation Computer Systems, 57, 42-55.
21. 21. Talebpour, M., Sykora, M., & Jackson, T. (2017, November). Social and Community Related Themes in Ontology Evaluation: Findings from an Interview Study. In International Joint Conference on Knowledge Discovery, Knowledge Engineering, and Knowledge Management (pp. 320-336). Springer, Cham.
22. 22. Verma, A. (2016, August). An abstract framework for ontology evaluation. In 2016 International Conference on Data Science and Engineering (ICDSE) (pp. 1-6). IEEE.
23. 23. Vrandečić, D. (2009). Ontology evaluation. In Handbook on ontologies (pp. 293-313). Springer, Berlin, Heidelberg.
24. 24. Xu, F. X., Liu, X. H., Chen, W., Zhou, C., & Cao, B. W. (2018). An Ontology and AHP Based Quality Evaluation Approach for Reuse Parts of End-of-Life Construction Machinery. Mathematical Problems in Engineering, 2018.