afuzzylogicsystemtoevaluatelevelsoftrust ... ·...

Revista Facultad de Ingenieriacutea No86 pp 40-53 2018

A fuzzy logic system to evaluate levels of truston linked open data resourcesUn sistema de loacutegica difusa para evaluar niveles de confianza en recursos de datos abiertosvinculadosPaulo Alonso Gaona-Garciacutea1 Jhon Francined Herrera-Cubides1 Jorge Ivaacuten Alonso-Echeverri1 Kevin AlexandreRiantildeo-Vargas1 Adriana Carolina Goacutemez-Acosta21Grupo de Investigacioacuten GIIRA Facultad de Ingenieriacutea Universidad Distrital Francisco Joseacute de Caldas Carrera 7 No 40B - 53 Bogotaacute DCC P 110231 Bogotaacute DC Colombia2Fundacioacuten San Mateo Transversal 17 Nordm 25 - 25 Bogotaacute C P 110231 Bogotaacute DC Colombia

ARTICLE INFOReceived September 212017Accepted January 12 2018

KEYWORDSLinked Open Data (LOD)Levels of Trust Open DataConsumption Fuzzy LogicFuzzy Systems Type I

Datos enlazados vinculados(LOD) Niveles de confianzaConsumo de datos abiertosLoacutegica Difusa SistemasDifusos tipo I

ABSTRACT Linked Data is a way of using the network by creating links among datafrom different information sources in order to improve search processes semanticinteroperability among other functionalities One of the strategies used to perform linkedresources queries is through the consumption of SPARQL Endpoints However to determineexistent trust levels within and outside the knowledge domains where such resources arethe operation of these Endpoints is one of the critical factors for both to realize the stateof these resources and their subsequent consumption To recognize these states thepresent article aims at exposing the description modeling implementation and analysis ofa type-I fuzzy system based on logical rules It also addresses decision-making regardingthe uncertainty that presents the definition of levels of trust to determine the consumptionobtained over a set of LOD Datasets Located in several Endpoints Finally it presentsresults conclusions and further work from a case study performed through the postulationof parameters obtained at runtime over several Endpoints

RESUMEN Linked Data es una forma de utilizar la red creando links entre datos de diferentesfuentes de informacioacuten con el propoacutesito de enriquecer los procesos de buacutesqueda lainteroperabilidad semaacutentica entre otras funcionalidades Una de las estrategias utilizadaspara llevar a cabo consultas de recursos vinculados es a traveacutes del consumo de EndpointsSPARQL No obstante el funcionamiento de estos Endpoints es uno de los factores criacuteticospara conocer el estado de estos recursos y su posterior consumo para determinar niveles deconfianza que presentan de estos recursos dentro y fuera de los dominios de conocimientodesde donde se encuentran alojados Para conocer estos estados el presente artiacuteculo tienecomo propoacutesito plantear la descripcioacuten modelamiento implementacioacuten y anaacutelisis de unsistema difuso de tipo-I basado en reglas loacutegicas estaacute asimismo orientado a la toma dedecisiones sobre la incertidumbre que presenta la definicioacuten de niveles de confianza paradeterminar el consumo que se tiene sobre un conjunto de Datasets LOD alojados en diversosEndpoints Finalmente presenta resultados conclusiones y trabajo futuro a partir de uncaso de estudio realizado a traveacutes de la postulacioacuten de paraacutemetros obtenidos en tiempo deejecucioacuten sobre varios Endpoints

1 Introduction

The evolution of the Semantic Web [1] has beenexperiencing a growing production of content on theWeb a scenario where a user can generate and publishresources that have or not a certain degree of validity

reliability and usefulness Likewise they can assume therole of consumer of content published by another personcalled the ldquoprosumerrdquo role [2] In order to carry out asuccessful process of linked data a series of rules onthe publication of web-based data has been proposed [1]These rules specify that the data has to be available in RDFformats and that to be queried prosumers need to usea standard SPARQL query language on a SPARQL Endpoint

Despite the existence of recommendations for thepublication of data there have been significant challenges

40

Corresponding author Jhon Francined Herrera Cubides

E-mail jfherreracudistritaleduco

ISSN 0120-6230

e-ISSN 2422-2844

DOI 1017533udearedinn86a0640

PA Gaona-Garciacutea et al Revista Facultad de Ingenieriacutea No 86 pp 40-53 2018

[3] in linking processes both at the metadata level and atthe data level itself From the study of these challengesone of the motivations that started this research isoriented to answer the following questions How do youadd confidence to the published resources How do youmeasure those rules of data publication in the Web frommechanisms that allow you to evaluate their effectivenessHow do you determine trust parameters for evaluatingthe veracity of the sources linked under principlesof Linked Data A scenario where the application ofthese dimensions is important for example it is inthe education domain where there are many availableeducational resources published on the Web as a resultthere is an increasing need of linking and discoveringthem Moreover due to this huge amount of educationalresources the need of establishing levels of confidencetrust and quality on such contents is evident

For this reason one of the relevant aspects is to determinethe concept of quality understood as the rdquofitness foruserdquo based on a proposal made by [4] Therefore thevalidation of the model proposed in this research willallow among other things to know the real state of thelinked data in the Web In addition to determine aspectsthat allow evaluating and defining if a Dataset (or datawithin a Dataset) can be discovered automatically and thepossibility of creating links between the data resources(Datasets and data elements)

There are some proposals aimed at adding confidence tothese resources with very variable information quality Infact different techniques have been used such as

bull PageRank [5] a method for objectively andmechanically rating webpages and thus effectivelymeasuring the human interest and attention devotedto them

bull Content-based trust [6] a trust judgement on aparticular piece of information or some specificcontent provided by an entity in a given context

Many of these techniques focus on the userrsquos contributiontrust links content of their metadata etc However thereis a concern about how to assess the levels of confidenceoffered by current resources specifically about how toassess the levels of trust offered by the current SPARQLendpoints that are available in the repositories whereseveral of them present problems as

bull Broken links [7] that do not allow reaching otherlinked resources

bull The resources have been relocated from the serverbut their references are not updated

bull Whenever an URI returns an error it is becausenon-reference issues arise

bull The server or SPARQL Endpoint is delayed or cannotrespond to a SPARQL query

This is how the SPARQL Endpoints that currently existshould be able to execute any SPARQL query Howeversome Endpoints are not able to resolve queries whoseexecution time or response cardinality exceeds a certainvalue while others simply stop execution withoutproducing any response In addition some limitationshave been found by Vidal et al [8] where the data accessibleon the Web are usually characterized by i) absence ofstatistics ii) unpredictable conditions when executingremote queries and iii) changing characteristics

Due to these situations this research seeks to addressthe SPARQL Endpoints consumption to evaluate thetrust they offer and it is especially focused on twocharacteristics of Accessibility Availability and ResponseTime Consequently this research is based on fuzzy logictechniques given that this technique

bull Plays an important role so that the margin of theclassification is more precise

bull Allows managing the level of uncertainty associatedwith the aggregate information of LOD

bull Allows optimizing the process of consumptionrecovery and evaluation of the trust offered by theconsulted Endpoints on these two measurementcriteria

In order to carry it out this article presents six sectionsIn section 2 a presentation of the background of themain references that define the research area is revealedSection 3 presents the proposed framework for theevaluation of trust levels and their articulation with FuzzyLogic Section 4 presents the results obtained Section5 discusses the results Finally section 6 presentsconclusions and further work

2 Background

The following section is a description of the main conceptsused in this research regarding levels of trust in theconsumption of SPARQL Endpoints

21 Linked open data data and metadata

Linked Open Data refers to a set of the best practices forthe publication and interconnection of structured data onthe Web in order to create a global interconnected dataspace called Web of Data [9]

This process of publication and interconnection ofdata requires an ensemble of information expressed in

41


data models As described by Ahmad et al [10] datarepositories expose information across multiple modelsA model must contain

bull The minimum amount of information transmittedto the consumer the nature and content of theirresources

bull Information enabling the discovery exploration andexploitation of data

This information is classified into the following types

bull General information General information about thedata set (eg title description ID) The data set ownermanually fills out this information In addition to thatlabels and domain information are required to classifyand improve Dataset detection

bull Access to information Information on access and useof the Dataset This includes the Dataset URL somelicense information (eg license title and URL) andinformation about Dataset Resources Each resourcegenerally has a set of attachedmetadata (for exampleresource name URL format size)

bull Property information Information about the Datasetproperty (eg details of the organization details ofwho provides support of the author) The existence ofthis information is important to identify the authorityin which the generated report will be sent and thenewly corrected profile

bull Source information Temporary and historicalinformation about the Dataset and its resources(for example creation and update dates versioninformation version number) Most of thisinformation can be automatically filled out andtracked

Data repositories are Dataset access points (endpoints)that provide tools to facilitate publication exchangesearch and display of data CKAN is the worldrsquos leadingplatform for managing open source data repositorieswhich makes websites more powerful such as theDatahubio that hosts the LOD cloud metadata

However what is a Dataset As described in [9] aDataset is any web resource that provides structured data

bull Tabular data data tables that can normally bedownloaded as CSV files or spreadsheets

bull Object Collections XML documents or JSON files

bull Linked data the standard web to describe the dataYou can find linked data in RDF resources or inSPARQL Endpoints

As Datasets are web resources they must have aURL Generally Datasets URLs are discovered in datarepositories or other data sources published by a dataprovider Often data sources publish links to Datasetsalong with their metadata which is structured data aboutthe Dataset itself (for example the Dataset author thelast Dataset update time etc)

However as the goal of this research is oriented tothe data published and beyond the metadata itself it isnecessary to take into account that the principles of LinkedData suggest the fulfillment of a series of levels wheretheir last instances are oriented to use standard openformats such as RDF and SPARQL in addition to link theirdata with the data of other people thus forming graphs ofknowledge [11]

Most providers that reach these levels provide a SPARQLEndpoint that in general terms is a SPARQL protocolservice as defined in the SPROT (The SPARQL Protocolfor RDF) specification A SPARQL Endpoint allows users(human or otherwise) to query a knowledge base throughthe SPARQL language Usually results return in oneor more machine-processable formats Therefore anEndpoint is mainly a machine-friendly interface to aknowledge base [12]

SPARQL Endpoints consumption can generate differenttypes of perceptions in the users given the existing trustlevel in that Endpoint If the respective links are not storedin these Datasets [8]

bull The query responses undertaken may be incompleteand important data may not have been included in theresponse

bull If it is required to evaluate the relationships betweentwo Datasets in order to determine possible newassociations these queries could be affected byreducing the level of trust in that Endpoint

Therefore the work of evaluating and verifying theinformation retrieved from a query can lead to twoscenarios the expected one contributing to obtainthe highest level of trust and the unexpected oneinconsistent or null decreasing the level of trust thatthe user can perceive on the Endpoint that is consumingEuropeanaLabs [13] describes an example of access andconsumption of SPARQL Endpoint and Kabutoya et al [14]provides an example of how to manage trust levels

22 Trust and data quality

As described by Gil et al [15] many authors havestudied the concept of trust in various areas of computingand in the context of the Web and the Semantic Web

42


Previous work on trust has focused on aspects such asreputation and authentication Trust is an important issuein distributed systems and security Many authenticationmechanisms have been developed to verify if an entityis which it claims to be usually using public and privatekeys Many access control mechanisms which are basedon policies and rules have been developed to be certainthat an entity can access specific resources (informationhosts etc) or perform specific operations Other authorshave also studied the detection of malicious or untrustedentities in a network traditionally in security and morerecently in P2P networks and e-commerce transactions

In the context of Linked Data Endpoints handle linksto other resources The Accessibility dimension (throughits metrics of availability and response times) allowsevaluating the trust offered by such Endpoint Thesemetrics allow us to obtain arguments to answer questionssuch as What SPARQL Endpoints can be trusted whenyou need to consult them [16] Question that guides theresearch process presented in this paper To address thisquestion the following aspects were raised

a Data quality is a multidimensional constructionwith a popular definition as the rdquofitness for userdquoData quality may depend on several factors suchas accuracy timeliness completeness relevanceobjectivity credibility comprehension consistencyconciseness availability and verifiability In termsof the Semantic Web there are several concepts ofdata quality Semantic metadata for example is animportant concept that must be taken into accountwhen evaluating the datasets quality But on theother hand the notion of link quality is anotherimportant aspect presented in Linked Data in which itis automatically detected if a link is useful or not [4]

b As stated by Zaveri et al [4] within the accessibilitydimension which involves aspects related to the waydata can be accessed and retrieved the followingmetrics can be found

bull Availability The availability of a Dataset isthe extent to which the information is presentobtainable and ready to be used

bull Response time It measures the delay usually inseconds between the display of a query by theuser and the receipt of the complete responsefrom the data set

c Since each of the SPARQL Endpoints that are availablehave their own navigation form and data exploration itwas considered important to use a graphical notationfor the development of this process

The LD-VOWL (Linked Data - Visual Notation forOWL Ontologies) was used to perform the graphical

representation based on the extraction of information fromthe Endpoint schema of linked data and its correspondingdisplay of the information using a specific notation[17] In [18ndash20] different studies are presented on thecharacteristics of VOWL and its comparison with othertools In brief there are different standpoints on VOWL

bull The VOWL notation allows a more compactrepresentation of ontology

bull Coloring the concepts property types and instanceshave a positive impact on task solution It helped toidentify elements and find information in the graphicalrepresentation

bull In Linked Data the LD-VOWL extracts schemainformation from Linked Data endpoints and displaysthe extracted information using the VOWL notation

bull SPARQL queries are used to infer the schemainformation from the endpoint data which is thengradually added to an interactive VOWL graphvisualization

The VOWL notation apparently provides only one possibleway to visualize OWL ontologies using node-link diagramsAs OWL is not an inherently visual language other typesof visualizations would also be possible and could be moreappropriate in certain cases For instance if users aremainly interested in the class hierarchy contained in anOWL ontology they might prefer a visualization that usesan intended tree or a treemap to depict the ontology

Some of the purposes for which this LD-VOWL proposalwas used are

bull To unify queries and simplify the process ofaccessing different Endpoints from the sameplace in addition to being able to include the variablesunder investigation

bull To work with a tool that manages a visual languageimplementation to allow the expression of thedifferent relationships and links between the entitiesinvolved in the queries of each user

bull This tool offers a friendly and understandablenavigation experience for the majority of Internetusers without knowledge in this field of work

Moreover authors such as Weise et al [21] Lohmann et al[22] and Saacutenchez [19] define this tool with a high level ofusability in aspects such as

bull Other tools can be readily integrated with LD-VOWLas it runs completely on the clientrsquos side and it onlyrequests the server through SPARQL for exampleWebVOWL [22] and Proteacutegeacute [22 23]

43


bull It is a well-specified visual language for therepresentation of user-oriented ontologies Thisdefines the graphical representations for mostelements of the Ontology Web Language (OWL)combined with a force-directed graphic design thatvisualizes the ontology

bull It manages interaction techniques that allow theexploration of the ontology and customization of thevisualization

Bearing this in mind along with the work undertakenwith the proposed Visual Data Web notation and incompliance with the specifications of the VOWL projectthe consumption of the classified and selected Endpointswas carried out according to the vocabularies allowedby VOWL which include RDF RDFS OWL and SKOSLikewise they were selected according to the strategiesthat each Endpoint uses to categorize and managethe linked resources Thus facilitating the necessaryinputs and factors of the conception of the diffuse logicmodel proposed for this research work and supportingthe construction and improvement of methods andmechanisms for both the collection and evaluation oflinked data on the Internet

The following sections present in detail the aspectsthat were taken into account for the use of this tool theprocess developed for the inclusion of the analysis ofvariables and the respective implementation of the Type-IFuzzy System based on logical rules which eventuallyallowed the evaluation of these variables

23 Fuzzy logic and the trust dimension inlinked data

Fuzzy logic is a methodology that provides a simple wayto get a conclusion from vague ambiguous inaccuratenoisy incomplete or with a high degree of imprecisionunlike conventional logic that works with well-defined andprecise information It also allows to be implemented inhardware and software or a combination of both [24]

This type of logic is capable of handling several typesof ambiguity Where x is a fuzzy set and u is a relatedobject the statement rdquou is a member of Ardquo is not alwayseither exactly true or exactly false It may be true only tosome degree the degree to which u is in fact a memberof x A crisp set is specified in such a way as to divideeverything under discussion into two groups membersand non-members A fuzzy set can be specified in amathematical form by assigning to each individual inthe universe of discourse a value giving its degree ofmembership in the fuzzy set [25]

In this context the use of fuzzy logic will allow abstracting

information about the trust offered by the Endpoints whichcome to offer information with a degree of inaccuracy Inthis domain different investigations have been identifiedsuch as those proposed by [26ndash28] It is important tonote that many of these proposals do not consider amore intuitive visualization tool for the deployment ofinformation resulting from queries On the other handits focus is on confidence in information annotationscontent and reputation In short in this research fuzzylogic is proposed to optimize the process of consumptionrecovery and evaluation of the trust deployed by Endpointsbased on the use of AccessibilityAvailability and ResponseTime dimensions in fuzzy logic rules

3 Methodology

This research is established in a quasi-experimental typemethodology where this design as an empirical methodallows performing analysis of properties resulting fromthe application of the technological process in order toobtain an analysis of the variables to work that is to saythe accessibility and the response times as dimensionsthat allow generating trust in the accessibility of the dataon the part of users

From this perspective a quasi-experimental designwas carried out to implement a strategy that providesinformation about aspects identified in the problempresented by Vidal et al [8] and Vandenbussche et al[16] information that will allow identifying the level oftrust in accessibility that these queries can throw whenconsuming such Endpoint In order to carry out thisproposal a structured methodological design is definedin phases which allows us to determine the processesleading to our research proposal (figure 1) In orderto carry out the quasi-experimental methodology themethodological design starts from a Preliminary Stage(inputs) where different exercises were carried out as a)connection to LOD repositories b) Endpoint Consumptionc) Theoretical review of the accessibility All of theforegoing as a strategy to obtain information about thelevels of trust offered by an Endpoint

The following are considered as subsequent stages

minus Download Configuration and Test of the SelectedVisualization Tool (VOWL) This stage considers thewiring process with the VOWL tool to establish theworking environment that will allow the interactionwith the designed system and the results from thequeries made

minus Identification and Design of the Strategy to Addressthe Trust Levels (Type-I Fuzzy Sets) This stageaddresses the resolution of the uncertainty generated

44


Figure 1 Proposed Methodological Design Source Authors

by the definition of levels of trust at the time ofconsuming linked data

minus Implementation and Testing of the ProposedFuzzy Model This stage describes the behaviorof the current system through its interaction withelements from the VOWL tool and with the necessarymechanisms to introduce the fuzzy logic model anddefines the definition of concepts on levels of trust

minus Generation and Analysis of Results Obtained Thisstage contemplates the process of examining theresults of implementing the proposed system anddescribes aspects of demonstrable levels of trustduring execution

4 Development of methodologicaldesign

41 Proposed framework

The proposed model (Figure 2) is composed of twofundamental structures In the first instance theframework which allows to obtain a display of theconsulting process to an Endpoint from the wiring with thetool VOWL and the available arguments at runtime In the

second instance it displays the creation of establishedrules for the declaration of levels of inference regardingdata consumption dimensions this will allow the systemto make a series of decisions as a result of the processof treatment of the parameters defined under the designand implementation of the Type-I Fuzzy System Thefollowing sections describe briefly the development ofthe proposed methodological design

42 Visualization tool VOWL

Based on the approaches defined by Lohmann et al[29] OWL does not have a standardized visual notationlike other related modeling languages However itsvisualization is very useful to be able to have a betterunderstanding of its structure and to be able to identifyin a more direct way those key elements to analyze inthe process of consuming Endpoints In this process ofsearching tools for the visualization of OWL [30] VOWLwasidentified like a complete tool which uses a well-specifiedeasy to understand and to implement notation

Under these criteria it is proposed to extend the VOWLfunctionalities [31] in order to analyze the suggestedmechanisms of trust levels to validate the developedinference rules as well as to determine a mechanismto obtain information about the times consumed in the

45


Figure 2 Proposed Framework

process of batch query of the resources exposed inthe Endpoint These aspects were considered key toknow the state of the Endpoints Therefore from theVOWL visualization strategies it is proposed to builda model based on fuzzy logic as regards VOWL thatallows determining the degree of performance that theapplication reaches for each of the queries made in termsof satisfactory requests and the number of seconds ittakes to obtain The above with the aim of improving thecurrent functionality provided by the tool and generatingmore elements of interest to Linked Data consumers

From these exposed functionalities according to thelevels of trust the architecture proposed on VOWLpresents two main factors

bull First the response time in each query andorconsumption made by a Linked Data user which ismeasured from the moment a request is executeduntil there is a response from the Endpoint whetherit contains or not associated data or until time runsout and no response has been obtained

bull Second it represents the return rate adjustedto the overall performance of consumption takingadvantage of the positive and negative results of theconsumption In this way it was possible to establishthe relationship with the dimensions and metrics thatare presented and worked in Zaveri et al [4] Inaddition the necessary inputs were provided for the

proposed Fussy Logic Model that is explained in thelater sections

43 Proposed model of fuzzy logic

As part of the proposed design it is established to workwith Type-I Fuzzy Systems [32] with a design of logicalrules based on two variables obtained from the querycomponent to the Endpoint the number of successfulrequests on the total requests in a query and the averageresponse time taken by each consultation round inresolving data consumption

These variables are structured in the form of antecedentswhich represent sets used to obtain a decisioncontemplated in the consequent sets The choice ofthis approach is based on two premises 1) in theenvironment of uncertainty generated when queryingDatasets and 2) the need to solve this uncertainty withcontrol over the operation of the system The abovethrough obtaining arguments at runtime and from theapplication to conceive the levels of trust associated witheach phase of implementation

For the implementation of the proposed model thefollowing procedures were defined

a Definition of variables The variables to be consideredfor the problem presented correspond to

46


bull NSR (Number of Successful Request) The valueof the proportion of requests that are successfulfrom the total number of requests at the time ofthe query

bull RT (Response Time) The time taken in secondsfor the application to respond to an associatedquery

b Declaration of fuzzy sets The fuzzy sets that areassociated with membership functions are established(Table 1) for the antecedents and the consequents ofthe system through the definition of concepts that willadjust the ambiguity with which the system counts tomake decisions about performance in execution time

c Attributes for fuzzy sets The attributes used (Table 2) aredefined for the definition of sigmoidal sets in terms ofantecedents and Gaussian in terms of the consequentones taking advantage of their characteristics ofcontinuity and derivability by analyzing the mostrepresentative values in each scenario of ambiguity andthe determination of the same ones for each function ofbelonging

d Definition of fuzzy sets The fuzzy sets are defined(Figure and ) in the dimensions of the antecedentsand the consequent ones by means of the conventionsand attributes previously defined in order to be able toimplement the system based on logical rules

e Design of logical rules The logical system design isestablished (Figure e) taking into account the use ofMamdanirsquos mathematical model [33] for the materialimplication of the logical conditional as given in Eq e

p rarr q equiv p and q (1)

Eq e Mamdanirsquos Implication

From equation e where p is the antecedent and qis the consequent figure e presents the rules of thesystem

f Design of the fuzzy systembased on logical rules In termsof the antecedents and consequents proposed (figure 5)in a fuzzy logic environment the system receives thesevariables as arguments and makes the decision whereappropriate in the sets described in the consequent

44 Implementation of the proposed model

As regards the design and implementation of theexperiments proposed for the model the followingaspects of the system were defined

minus Discretization or postulation of values

minus Fuzzification using singleton

minus Aggregation of rules using the Mamdanirsquos Model

minus Defuzzification with weighted center-average orcentroid

minus Link to functional components

However taking into account the use of the antecedentsin the system to grant a decision which in this case is toprovide an associated performance value aspects of theconsequent that are being handled from the design wereconsidered during the implementation phase

bull Optimal Performance (OP) It occurs in the bestpossible execution scenario (from 70 to 100)This performance is the ideal scenario in an LODresource consumption environment considering thatthe query result has the expected values of the definedvariables This index may not reach full performancegiven the proposal to work with fuzzy systems but itallows contemplating all those scenarios in which thequery has an excellent level of trust

bull Average Performance (AP) It corresponds to themost standard scenario of the implementation withinan environment of uncertainty (from 30 to 70)This performance scenario is the most known whenconsuming LOD resources It denotes the need todefine levels of trust for consumption in addition tosome characteristics of the uncertain environmentthat this environment presents

bull Deficient Performance (DP) It is a frequent scenariobut it shows an unwanted situation of a query (from0 to 30) This scenario represents the averageperformance since its results are unattractive whenconsuming LOD resources hence the implementationof the Fuzzy Logic Model from this point couldconsider representative faults on the resources andit could make suggestions on the organizations thatare in the provision of these queries

bull Critical Performance (CP) It is the worst possiblescenario of execution (0) This is the worst caseobtained from a consumption of LOD resources sinceits results do not produce any synonym of successConsequently a series of imminent failures haveto be fixed insofar as the decisions are made frominference rules to ensure a successful consumptionof resources

After carrying out the corresponding implementation of thefuzzy sets and the respective rules concerning the VOWLvisualization tool [31] services were used in order to firstcarry out a count of requests on Endpoints second take acount of the number of relationships and third calculatethe response time of these Endpoints from the averagetime between each query request These aspects areintegrated for determining the consumption of linked datafrom an available Endpoint

47


Table 1 Membership Functions

mNSR-HIGH Membership Function of the NSR Concept of HighmNSR-LOW Membership Function of the NSR Concept of HighmRT-HIGH Membership Function of the RT Concept of HighmRT-LOW Membership Function of the RT Concept of LowmOP Membership Function of the Optimal PerformancemAP Membership Function of the Average PerformancemDP Membership Function of the Deficient PerformancemCP Membership Function of the Critical Performance

Table 2 Fuzzy Sets Attributes

mNSR-HIGH Center 50 mNSR-HIGH Dispersion 60mNSR-LOW Center 50 mNSR-LOW Dispersion 60mRT-HIGH Center 40 mRT-HIGH Dispersion 1mRT-LOW Center 30 mRT-LOW Dispersion 1

mOP Center 1 mOP Dispersion 01mAP Center 07 mAP Dispersion 01mDP Center 03 mDP Dispersion 01mCP Center 0 mCP Dispersion 01

5 Results analysis

There are a number of research related to the formulationand determination of methodologies aimed at ensuringthe quality of the data consumed through which it ispossible to postulate dimensions such as accessibilityand within which the availability and response time of thedata are identified [4] Thus the development of a systembased on inference rules by defining fuzzy sets withconcept formulation under the main dimensions of qualitydata is a research with practical implementation thattranscends the methodologies oriented to the analysisof data consumption In the concept of Open Data thisalso provides a dedicated development to the resolution ofconflicts when acquiring information

Evaluating the availability criteria and response timeof the Endpoints in a practical scenario allows observingthe behavior of the linked open data through an automateddevelopment that works with methodologies of qualityassurance in addition to offering new characteristicsin the context of the consumption of the information inEndpoints of a dataset

During the system execution phase derived from theseconsumptions figure 6 shows the obtained results whichshows the display of regular VOWL visualization alongwith the collection of the necessary information for theoperation of the designed fussy system (right side of theimage) Figure 7 shows the components offered by VOWL

bull URL it corresponds to the endpoint consumed

bull Pending Requests it shows the number of requests

pending for completion within the data consumptionprocess

bull Successful Requests it displays the number ofcompleted requests within the data consumptionprocess of the current query

bull Failed Requests it is the number of failed requestswithin the data consumption process of the currentquery

Together with the results obtained by the insertion of fuzzysets (Figure 7)

bull Times it displays the average time among queryrounds In the business layer this time establishesthe metric the system manages to adjust the conceptunder the corresponding set to the response timesand describes the mechanism the argument uses tointeract with the inference rules to produce a resultwhich translates into a performance index of the levelof trust obtained

bull Performance it displays the performance index ofeach query which is the result of the system basedon inference rules after having worked with thearguments obtained from the number of successfulrequests and the response time in a presentationlayer format that translates levels of trust processedduring data consumption into a metric indicator onquality data

As seen in the queries performed where Endpoints witha small amount of data were consumed the Performanceindicator (as analysis between complete requests versustime) varies which establishes that the higher the

48


(a) Definition of diffuse set antecedents (b) Definition of diffuse set Consequents

Figure 3 Definition of diffuse set

Figure 4 Design of logical rules

percentage value is the higher is also the level of trustoffered by the Endpoint In this section it is importantto highlight some aspects to consider because theseinfluence the performance of the query

bull Endpoints with large volumes of data their responsetime may be high but reliable

bull Performance is influenced by how the platform isimplemented where the Endpoint is and by theContent Provider being part or not of its entiredata-binding platform

To validate the performance additional executions werecarried out with different SPARQL endpoints in a randomway (Figure 8) to obtain detailed information in percentageterms about the requests completed (return rate) Thisallows defining the trust levels in each of them accordingto the criteria defined in the model in the sense thatnow the system recognizes the trace of ambiguity whichexists when consuming data on each Endpoint andcan determine the state of the Endpoint based on thedimensions described by the concepts of fuzzy setsFinally according to the implementation of the fuzzysystem designed for the application the surface graph(Figure 9) shows how the interaction between the twoinput arguments of the dataset query emits a result thatis translated in terms of associated performance to theconsultation rounds

It is thus that Figure 9 allows a more in-depthunderstanding of the interior of the proposed fuzzysystem and how at runtime the possible argumentsused would yield a contemplated response in the designof the ensuing sets which is reflected in terms of theperformance of the query to the requested dataset

6 Discussions

As can be seen in the results obtained the management ofresponse time and availability metrics provide informationon how the trust offered by published SPARQL endpointscan be evaluated The performance obtained in theEndpoint consumption applying the fuzzy sets allows toclassify them in levels of trust proposed in this researchThis process contributes to obtain a differentiation in theexperience that the user lives in the consumption of linkedresources

In this context the response time and availabilityattributes can be used [4] to determine levels of trustin the Endpoint consumption The main objective of ourframework contributes to the aim proposed in [31] rdquoTheneed for new developments based on specific analyzes ofdatasets to determine the most appropriate and practicalbasis for creating a comprehensive conceptual frameworkfor data qualityrdquo

Similarly these authors [31] state

bull Availability such as the ease of locating andobtaining an information object (for exampledatasets) a dimension for which metrics such asserver accessibility SPARQL Endpoint accessibilitydereferenciability problems and others can be used

bull Response time such as the amount of time until thecomplete response reaches the user Dimension forwhich the percentage of response in time (per second)can be measured

Thus determining the tacit importance of relating thenumber of successful andor completed requests alongwith the time spent for each of them making possible thecreation of the Type-I fuzzy logic system presented in thisresearch

Therefore the obtained results (Figure 10) are alignedwith the indicators and metrics proposed in [34] insofaras the performance achieved by each SPARQL Endpointis understood as optimal as a function of a) a time offast response and b) the resolution of a high number

49


Figure 5 Design of the fuzzy system based on logical rules

Figure 6 VOWL display with fuzzy logic component

Figure 7 Endpoint consumption results

50


Figure 8 Percentage of successful requests

Figure 9 Surface Graph Interaction Arguments

Figure 10 Resultant Variables

of requests according to how the functions of belongingwere conceived for the established fuzzy logic systemIn this way a favorable index for the search proposal oflevels of trust was obtained and it is presented in thisresearch work The above arguments can be evidenced inthe research exposed in [34] where it is detailed that

bull The higher the accessibility metric is it is better (it isrecommended gt 1 req psec)

bull The lower the response timemetric is it is better (lt05seconds per request)

Metrics that when contrasted allow obtaining thecategorization of the rdquoPerformancerdquo proposed in thisresearch (Optimal Acceptable Poor and Critical)

Finally in order to compare this approach with otherdifferent investigations like [21 22] have used VOWLto expose a uniform visual notation for OWL ontologiesbased on aspects such as clarity and distinguishability ofelements ease of use with interactive highlighting anddesign features as well as aesthetics

On the other hand the implementation of fuzzy logictechniques has made it possible to use the informationprovided by the Endpoint queries and to be able toestablish classification criteria that allows evaluating thelevel of trust offered by each Endpoint In proposals suchas [25 26 35] the visualization of the information resultingfrom the Endpoint Consumption is observed Also thesuggestion for future works regarding the inclusion of newquality criteria that will allow evaluating the trust offeredby such Endpoints

In conjunction with the above it becomes apparentthat the application of fuzzy sets for the evaluation of themetrics of availability and response time in addition toaligning with the proposed theoretical references offers

51


a development based on specific analyses of datasetsThese fuzzy sets allow determining appropriate andpractical criteria to provide a guidelines of reliability in theconsumption of data and therefore in data quality

7 Conclusions

The process of adding trust to published resources startsfrom an adequate clear and public abstraction of a datamodel Subsequently monitoring the recommendationsdescribed to carry out the exposure and linkage processcontributes to the fact that the published data offers bothgreater usability and a higher value of trust to the user

The above can be seen in aspects such as the datahave metadata its resources have an identifier and canbe referenced the data are published in suitable formatsthey can be reused and the platforms where they arestored offer services for query among other factors All ofthis contributes to the construction of trust by those whoaspire to consume those resources

In addition evaluating different quality dimensionsoffering consumers relevant mechanisms and indicatorsof these dimensions resulting from the application ofthese mechanisms contributes to the building of trustin the resources exposed Such is the case of indicatorslike availability and response time indicators that allowevaluating if a resource is available and how long ittakes to resolve its referencing These factors allow theconsumers to build trust by giving them information aboutwhether the resource they want to access is availablereferenceable and resolved in a relatively acceptable time

As a result the Type-I Fuzzy Systems were selected asmechanisms to evaluate the proposed quality dimensionswith a design of fuzzy logic rules based on two variablesThese variables were obtained from the query componentto the Endpoint the number of successful requests onthe total number of requests in a query and the averageresponse time taken by each query round in resolving dataconsumption

This mechanism coupled with a visualization tool suchas VOWL allowed evidencing and analyze the rdquoTimerdquo andrdquoPerformancerdquo indexes associated to data consumptionand included in this research which provide informationabout the response time and performance level obtainedcategorizing in this way the level of trust (OptimalAcceptable Poor Critical) obtained in the consumptionof such resources Therefore through the proposedapproach the fuzzy aspects immersed in a process ofdata consumption were conceived in order to take thesefeedbacks and contemplate a more horizontal scenario inthis area

In general the design and implementation of a Type-I fussy system was projected based on logical rules for the definition of trust levels as a preliminary model that allows a more robust control of the data recovery in the consumption processes of the Endpoint

On the other hand the use of such sets opens the possibility of constructing more proposals or models more rigorous ones by adding different classifications of each of the schemas and datasets optimizing and greatly improving the levels of trust which will increasingly favor the quality of use and consumption of Linked Data However it is necessary to consider other work scenarios in other repositories and SPARQL connection points to define rules and methods that encompass and collect more schemas and resources in the Linked Data framework

As further work it is proposed to include more elements that feed the fuzzy system to achieve a much more detailed analysis of each of the schemas available in the linked data framework In this context it is intended to gather more dimensions in the framework of computational learning which allows defining levels of trust from more rigorous processes Moreover it is intended to reduce the uncertainty about the response time of Endpoints with a greater coverage of variables On the other hand future solutions might consider parameters that could be inferred in the queries while being able to increase the feasibility and trust for example in broken links

8 Acknowledgement

This research has been developed within the framework of the doctoral research project on Linked Data at the District University Francisco Joseacute de Caldas In the same way the subject matter is working as a line of Research Group GIIRA

9 References

[1] T Berners J Hendler and O Lassila rdquoThe SemanticWebrdquo ScientificAmerican vol 284 no 5 pp 29-37 2001

[2] A A Rodriacuteguez rdquoLas nuevas pautas para el acceso a la informacioacutenrdquoInvestigacioacuten Bibliotecoloacutegica Archivonomiacutea Bibliotecologiacutea eInformacioacuten vol 30 no 69 pp 117-135 2016

[3] W3C Data on the Web Best Practices 2017 [Online] Availablehttpsw3cgithubiodwbpbphtml Accessed on July 252017

[4] A Zaveri et al rdquoQuality Assessment Methodologies for Linked OpenData A Systematic Literature Review and Conceptual FrameworkrdquoIO Press Journal no 1 pp 1-5 2012

[5] L Page S Brin R Motwani and T Winograd rdquoThe PageRankcitation ranking Bringing order to the webrdquo Stanford UniversityStanford CA Tech Rep Jan 1998

52


[6] J Pattanaphanchai rdquoDC Proposal Evaluating Trustworthiness ofWeb Content Using Semantic Web Technologiesrdquo in The SemanticWeb ISWC 10th International Semantic Web Conference BerlinHeidelberg 2011 pp 325-332

[7] E Rajabi S Sanchez andM A Sicilia rdquoAnalyzing broken links on theweb of data An experiment with DBpediardquo Journal of the Associationfor Information Science and Technology vol 65 no 8 pp 1721-17272014

[8] M E Vidal et al rdquoRetos para el Procesamiento Semaacutentico de DatosEnlazados en la Nube de los Datos Abiertos Enlazadosrdquo RevistaVenezolana de Computacioacuten vol 1 no 1 pp 34-42 2014

[9] LinkedDataCenter What is a Dataset 2017 [Online] Availablehttpsiteslinkeddatacenterhelpbusinessstarter-kitdataset Accessed on July 20 2017

[10] A Ahmad R Troncy and A Senart rdquoWhat is up LOD CloudObserving the State of Linked Open Data Cloud Metadatardquo inInternational Semantic Web Conference Portorož Slovenia 2015 pp247-254

[11] T Berners Linked Data 2009 [Online] Available httpswwww3orgDesignIssuesLinkedDatahtml Accessed on May20 2017

[12] Semantic WebSPARQL Endpoint 2011 [Online] Available httpsemanticweborgwikiSPARQL_Endpointhtml Accessedon June 20 2017

[13] Europeana pro Europeana SPARQL Endpoint 2016[Online] Available httpsproeuropeanaeuposteuropeana-sparql-endpoint Accessed on July 12 2017

[14] Y Kabutoya R Sumi T Iwata T Uchiyama and T Uchiyama rdquoA Topic Model for Recommending Movies via LinkedOpen Datardquo in IEEEWICACM International Conferences on WebIntelligence and Intelligent Agent Technology Macau China 2012 pp 625-630

[15] Y Gil and D Artz rdquoTowards content trust of web resourcesrdquo Web Semantics Science Services and Agents on the World Wide Web vol 5 no 4 pp 227-239 2007

[16] P Vandenbussche A Hogan J Umbrich and C Buil rdquoSPARQL AGateway to Open Data on the Webrdquo ERCIM NEWS Linked Open Data vol 1 no 96 pp 31-31 2014

[17] VisualDataWeb LD-VOWL Visualizing Linked Data Endpoints 2016 [Online] Available httpvowlvisualdataweborgldvowlhtml Accessed on May 16 2017

[18] S Lohmann F Haag and S Negru Towards a Visual Notation for OWL A Short Summary of VOWL 2013 [Online]Available httpspdfssemanticscholarorg2c26fdeac23c325f2832ed786a0beba656605522pdf Accessed onJune 28 2017

[19] AIMS Agricultural Information Management Standards VOWLVisual Notation for OWL Ontologies 2014 [Online] Availablehttpaimsfaoorgcommunitygeneral-informationblogsvowl-visual-notation-owl-ontologies Accessedon July 8 2017

[20] T Ramdane and S Bachir OWL Visual Notation and Editor 2014[Online] Available httpwwwuniv-tebessadzfichiersouarglaicaiit2014_submission_217pdf Accessed on July9 2017

[21] M Weise S Lohmann and F Haag rdquoLD-VOWL Extracting and Visualizing Schema Information for Linked Datardquo in 2nd InternationalWorkshop on Visualization and Interaction for Ontologies and LinkedData Kobe Japoacuten 2016 pp 120-127

[22] S Lohmann S Negru F Haag and T Ertl Visualizing Ontologies with VOWL [Online] Available httpwwwsemantic-web-journalnetsystemfilesswj1114pdfAccesed on July 23 2017

[23] D Bold S Lohmann and S Negru Proteacutegeacute VOWL VOWLPlugin for Proteacutegeacute 2014 [Online] Available httpvowl visualdataweborgprotegevowlhtml Accessed on July 232017

[24] M D Rojas E Zuluaga and J Ochoa rdquoPropuesta de medicioacuten de laconfianza en la informacioacuten utilizando un sistema difusordquo RevistaIngenieriacuteas Universidad de Medelliacuten vol 10 no 19 pp 113-123 2011

[25] S Javanmardi M Shojafar S Shariatmadari and S S Ahrabi rdquoFR

TRUST A Fuzzy Reputation Based Model for Trust Managementin Semantic P2P Gridsrdquo International Journal of Grid and UtilityComputing vol 6 no 1 pp 57-66 2014

[26] I Jacobi L Kagal and A Khandelwal rdquoRule-Based TrustAssessment on the Semantic Webrdquo in International Workshop onRules and Rule Markup Languages for the Semantic Web BarcelonaSpain 2011 pp 227-241

[27] S Kamal S F Nimmy and L Chowdhury rdquoFuzzy Logic overOntological Annotation and Classification for Spatial FeatureExtractionrdquo International Journal of Computer Applications vol 41no 6 pp 18-22 2012

[28] R Aringhieri E Damiani S De Capitani S Paraboschi and PSamarati rdquoFuzzy Techniques for Trust and Reputation Managementin Anonymous Peer-to-Peer Systemsrdquo Journal of the AmericanSociety for Information Science and Technology vol 57 no 4 pp528-573 2006

[29] S Lohmann F Haag and S Negru rdquoTowards a Visual Notation forOWL A Brief Summary of VOWLrdquo in International Experiences andDirections Workshop on OWL Bologna Italy 2016 pp 143-153

[30] W3C OWL Web Ontology Language Overview 2004 [Online]Available httpswwww3orgTRowl-features Accessedon July 25 2017

[31] 31 GitHub Inc LD-VOWL 2016 [Online] Available httpsgithubcomVisualDataWebLD-VOWL Accessed on April 252017

[32] J M Mendel rdquoFuzzy logic systems for engineering a tutorialrdquoProceedings of the IEEE vol 83 no3 pp 345-377 1995

[33] A Peregriacuten rdquoIntegracioacuten de Operadores de Implicacioacuten yMeacutetodos de Defuzzificacion en sistemas Basados en ReglasDifusas Implementacioacuten anaacutelisis y Caracterizacioacutenrdquo MS thesisUniversidad de Granada Granada Espantildea 2000

[34] C Bizer Z Miklos J Calbimonte A Moraru and G FlourisD21 Conceptual model and best practices for high-quality metadatapublishing 2012 [Online] Available httpplanet-datawikisti2atuploadsarchivedd72012021716035221D21pdf Accessed on May l5 2017

[35] VOILA 2016 Visualization and Interaction for Ontologies and LinkedData Visualization for Ontology Evolution 2016 [Online] Availablehttpvoila2016visualdataweborgVOILA2016_proceedingspdf Accessed on May 30 2017

53


[3] in linking processes both at the metadata level and atthe data level itself From the study of these challengesone of the motivations that started this research isoriented to answer the following questions How do youadd confidence to the published resources How do youmeasure those rules of data publication in the Web frommechanisms that allow you to evaluate their effectivenessHow do you determine trust parameters for evaluatingthe veracity of the sources linked under principlesof Linked Data A scenario where the application ofthese dimensions is important for example it is inthe education domain where there are many availableeducational resources published on the Web as a resultthere is an increasing need of linking and discoveringthem Moreover due to this huge amount of educationalresources the need of establishing levels of confidencetrust and quality on such contents is evident

For this reason one of the relevant aspects is to determinethe concept of quality understood as the rdquofitness foruserdquo based on a proposal made by [4] Therefore thevalidation of the model proposed in this research willallow among other things to know the real state of thelinked data in the Web In addition to determine aspectsthat allow evaluating and defining if a Dataset (or datawithin a Dataset) can be discovered automatically and thepossibility of creating links between the data resources(Datasets and data elements)

There are some proposals aimed at adding confidence tothese resources with very variable information quality Infact different techniques have been used such as

bull PageRank [5] a method for objectively andmechanically rating webpages and thus effectivelymeasuring the human interest and attention devotedto them

bull Content-based trust [6] a trust judgement on aparticular piece of information or some specificcontent provided by an entity in a given context

Many of these techniques focus on the userrsquos contributiontrust links content of their metadata etc However thereis a concern about how to assess the levels of confidenceoffered by current resources specifically about how toassess the levels of trust offered by the current SPARQLendpoints that are available in the repositories whereseveral of them present problems as

bull Broken links [7] that do not allow reaching otherlinked resources

bull The resources have been relocated from the serverbut their references are not updated

bull Whenever an URI returns an error it is becausenon-reference issues arise

bull The server or SPARQL Endpoint is delayed or cannotrespond to a SPARQL query

This is how the SPARQL Endpoints that currently existshould be able to execute any SPARQL query Howeversome Endpoints are not able to resolve queries whoseexecution time or response cardinality exceeds a certainvalue while others simply stop execution withoutproducing any response In addition some limitationshave been found by Vidal et al [8] where the data accessibleon the Web are usually characterized by i) absence ofstatistics ii) unpredictable conditions when executingremote queries and iii) changing characteristics

Due to these situations this research seeks to addressthe SPARQL Endpoints consumption to evaluate thetrust they offer and it is especially focused on twocharacteristics of Accessibility Availability and ResponseTime Consequently this research is based on fuzzy logictechniques given that this technique

bull Plays an important role so that the margin of theclassification is more precise

bull Allows managing the level of uncertainty associatedwith the aggregate information of LOD

bull Allows optimizing the process of consumptionrecovery and evaluation of the trust offered by theconsulted Endpoints on these two measurementcriteria

In order to carry it out this article presents six sectionsIn section 2 a presentation of the background of themain references that define the research area is revealedSection 3 presents the proposed framework for theevaluation of trust levels and their articulation with FuzzyLogic Section 4 presents the results obtained Section5 discusses the results Finally section 6 presentsconclusions and further work

2 Background

The following section is a description of the main conceptsused in this research regarding levels of trust in theconsumption of SPARQL Endpoints

21 Linked open data data and metadata

Linked Open Data refers to a set of the best practices forthe publication and interconnection of structured data onthe Web in order to create a global interconnected dataspace called Web of Data [9]

This process of publication and interconnection ofdata requires an ensemble of information expressed in

41
























42






















43











3 Methodology






44













45













46

























47







5 Results analysis













48










6 Discussions








49





50












51



7 Conclusions









8 Acknowledgement


9 References






52

































53
























42






















43











3 Methodology






44













45













46

























47







5 Results analysis













48










6 Discussions








49





50












51



7 Conclusions









8 Acknowledgement


9 References






52

































53






















43











3 Methodology






44













45













46

























47







5 Results analysis













48










6 Discussions








49





50












51



7 Conclusions









8 Acknowledgement


9 References






52

































53











3 Methodology






44













45













46

























47







5 Results analysis













48










6 Discussions








49





50












51



7 Conclusions









8 Acknowledgement


9 References






52

































53













45













46

























47







5 Results analysis













48










6 Discussions








49





50












51



7 Conclusions









8 Acknowledgement


9 References






52

































53













46

























47







5 Results analysis













48










6 Discussions








49





50












51



7 Conclusions









8 Acknowledgement


9 References






52

































53

























47







5 Results analysis













48










6 Discussions








49





50












51



7 Conclusions









8 Acknowledgement


9 References






52

































53







5 Results analysis













48










6 Discussions








49





50












51



7 Conclusions









8 Acknowledgement


9 References






52

































53










6 Discussions








49





50












51



7 Conclusions









8 Acknowledgement


9 References






52

































53





50












51



7 Conclusions









8 Acknowledgement


9 References






52

































53












51



7 Conclusions









8 Acknowledgement


9 References






52

































53



7 Conclusions









8 Acknowledgement


9 References






52

































53

































53

afuzzylogicsystemtoevaluatelevelsoftrust ... ·...

Documents