Embed Size (px)
Citation preview
Augmented Systems in-Ontology BasedIntelligentl nformation Retrieval
Poonam Yadav, Ravindra Pratap Singh . I
Abstract: In order to solve the problem of information overload, current Information retrieval systems need to. be irnproved , Intelligencoshould he embedded to. search systems to manage effectively search, retrieval. Filtering and presenting relevant information. This can bkdone by onlolo.gy based information retrieval techniques. Systems ;;Ire evaluated by using their information retrieval features.' Ontologietare introduced to. describe the semantic 'information of knowledge leases in order to. meet requirement of utmQS',tsharing and reuse of th~domain knowledge .The paper provides an overview of ontology-based informatiofl retrieval techniques and software tools currentltavailable as prototypes or commercial products.
2, INTELLIGENT AGENT TECHNOLOGY
.At present, Agent \videly lk"1s applied in the Hehir. ofinformation retrieval and service,. electr9nk commerce,distributed computing and so on. Yi Xiao. et al. des~ribedAgent 'is a concept developed in the field of the arnfici;llintelligence (AT), is a hot spot in ATt.he recent years. G~nel'ally
Keywords: Ontologies, Information Jetrieval, Semantic indexing, Semantic mapping .
• ---------- Idocument; these keywords can be used in docLnlentclassification which is del'elldent on ontology know 1edlge. Soinformation retrieval system can sort documents to~somepossible domains combinilig with prim(H}" cont~lt ofinformation docurnent, then filtrates off irrdevant dobulins,gets final result of corrc-Iative domains v.:hich the dOCUl~entis
Iclassed to. '1
.Moreover, it can be make keyword matching based m theapproximate semantic network of ontology, m.ltom ticalhclas::icorrelative document.s to .;1 certain d0111ain.fn thib ',,\";-1)',
information retrieval system searches some seleclivedocuments which are cor;elative to user searching de~Hi.md,and do no longer scan all infonnation dOCU1?len~s.Sosearching time and cost can be compressed greatly and!user'ssemantic relativity judgment may be expedited. i(2) Intelligent formulation and yisu(llization t.o luser"ssearching demand.For the searching keywords given by user, we can effectively.estimate feasible domain depending on ontology knoV\lledge.Then information retrieval system enumerate con~istE'ntconcept and definition of the domain to user, user can! makejudgment and selection. On the one hand, it can help userconfiril1 their knowledge demand, 1JEercan more disbnctJt;-visualize 'information demand which is not clearly presEnte.d
or little doe~ uSer think. On the Othl~l' hand, it c~.n It'tinformMion retric,;ai sl-stem ascertain the positi~n ofsearching key,"",ord in - ontology, so help mac1-uhe 1';)wlderstand user's siC'arching intelltion, and provide Illl()reaccurate and more relevant information. I
Because of ontology defines essential kennel concept andtheir semantic relation, it can be used as the tools of re~ourcedescription and knowledge presentation.
1. ONTOLOGY AND INFORMATION RETRIEVAL
JING-YAN et aL described that characteristics .of an idealinfor.!11ati(lllretrieval systems are that searching is fleet andresult is accurate. Traditional methods of information ret.rievalcan be divided into three essential kinds: text retrieval, dataretrieval and topic retrieval as described in [1].
Tt,:\t retrieval system: The user's information retrievaldeJllil.nd is represented itS keywords which are compared witheach word 'of whate text in text retrieval system. Its searchingmethod is based on word-frequency analysis, and typicalsystem is Google. The excellences of this kind of informationretriewd s:ystem are includes a large amount of outPut resultsand super automatic search control. Its shortcomings are thatreturn information is accumulated and reality is,too low.
Data- retrieval system aims at the structUral.informationsystem, retrieval demand and information data all stands bycertain formats, and is provided with the decided structure. Itpermits appointed field search, and various kinds ofcOlllm:ercial service database all adopt this searching nlOde.Tht, quality of data reh'ieval depends OIl code urganization,lnd rnay cost a .lot of money. The output result is relativeaccurate, but can not generate all expectancy i.nformation.
Topic retri.eval system colleds information by manpowermamwr or. semi-automatic manner, It scans accessingdOl'u.ment b); dedicated method, ilnd append. to somebeforehand determinate topk. User can visit dOWn\VCll'dsgradually from one layer to next layer, 50 finally can gelretrieval result. The representative system is Yal10(J..TIle excellence of topic retrieva'l system is that user C311
easily master searching process. Ontology is the .for~:nalizeddescription of share c0l1ceptual model to the informationresource, so any conceptua 1object C can be defined as:
C~{D, W,RjYVhere,.Dindicates such domain knowledge, W is a status
set of correlative thing in applied domain, R is a set. of conceptrelation in donlain .space {D"V"}: .. l~ has hvo effects that (;mtology' is used in information
reh'ieval.(1) Automatic analysis to the domain attribute of document.The info.rmation retrieval system looks for keywordsaccording- to a certain means in searched information
Fig.1 [2]: The. basic model of Agent system
The basic modei of Agent system generallY, there are threemost basic- properties of Agent such as the autono,my, learningand the cooperation. A basic Agent system should has One of .these 'properties, a advanced Agent system must have theautonomv and one of other two items, but an ideal Agentsystem n~ust 'realize al] of tliese properties and introduce other'charatteristics (for example mobility according to the actualdeI,n<:tnd,. etc).
3, CG INTELLIGENT INFORMATION RETRIEVAL SYSTEMARCHITECTURE
Users iilpllt keywords which .they want to search, thenretrieval system retu~n the matching document to users.B,ecause of synonyms an4 polysemy, it's very difficult totp:1derstand user exact needs by keyword. Often the initialkeywords do 1LOtget tiLe results they want, the mildly relevantor irrele~'ant documents were also retrieved. Pan Ying et a1described that CG introduces hO\\I to use the computer togenerate, process and display graphics. CG is vei.y broad in itscontent. It is difficult for learners to understand the wholesrstem of disciplines as qe~cribed in [3].
I L'",,'r r\;;'l~~'Y;li \1!1,-.II~j~'C
, I
<..:i 1<.,':1U :di,'h., ; )" •.,,~t1!'~",UUt'Sl t, ..~.~.,---~.~----~, _.-+~~-r.;f,~i-i:r,:::~';;~-r-c;Ur1;:--,-~''''''''C:' :ud ••~... ~
[~~"":~~ a' "PP"!~~~:~~jInvokel I
.•• ,A11n:~yr],,--,..;:1t1 J:.c:::;.;uJt,s,.
Fig 2[3]: CG Information Retrieval System architecture
On the client side, users input q'l.leries in the re~ri~va]interface. On the server side, retrieval application .acceJits therequest of users, and then invokE's Jena API to analys~s CGontologv ,md Forms search results, the final results 2lrereh,med to the users at last... I4. INFORMATION RETRIEVAL AGENT SYSTEM BASED ON
KEYWORDS DISTRIBUTION 1Jae-VVoo'LEE described that for efficient information ret ieval,it is important that keywords are defined very wbll <'IS
appropriate terms about all documents in. the 'lnte~1et ordistributed computing systems. And there are three, kU1ldsofalgClrithms, retrIeval algorithms, filtering algorithm! <mdindexing algorithms. Retrieval algorithms are extrbctingimportant or meanim"l'flll information froni 'dathba5e,
a TI . 'f' I ,knowledgebase or documents. 10se are ClaSSlleLL tosequential SGl1U1ingand searching indexed text files. Fi~eringalgorithms are simplifying result of text or informati~n forefficient information rettieval. Indexing algoritlunk [Ireconstr.uctil~g daia structure for s~arching information et~~uy~Indexmg IS, nne of the lI'lOSt nnportanl data for efflClent
I
information retrievaL For automated indexing lheTLe arevarious algorithms, .stemming, st~p Jist, thesalJTus, ~tc i'lS
discussed in [4]' IStemming is the process of matching morphological term
variants for using general terms. Stemming is for reducii1g thesize if index files and ,improving inforn~ation retrieya 1sy~teri1's
Inquiry system is B/S model, Java progrmTI which invokesJena APT functions to realize the complex Sernantic inqhiries,.Jena is a Java framework for building Semantic} vVebapplications. It provides a programmatic environment for RDF(Resource Description Fl'arnework); RDFS (RDF Schema) andOWL, SPARQL (Simple Protocol and RDFQuery Langltage),and includes a rule-based inference engine. lena can be used tocreate and manipulate RDF graFhs. Tl1e system n~ainlyincludes: th.e user retrieval interface, the retrieval appJj(~ation,Jena Ontology analytical tools, and the CG ontology. TheSystem archilecture is shown in fig.2.
2.
3.
4.
5.
believed that, Agent is the, intelligent entity which tan runindependently find continuously under the environment ofisomerism computation, and has the properties as follows asdescribed in [2J.
], Autonomy - Agent is able to run. continuouslywithout the disturbance from, outside or nobody, andcan control its own behavior and internal condition.Adaptability Agent can sense the externalenvironment, and carries all. the sel£- .adjustl1'l.entaccording to the change of -exteni.af environment,therefore has the ability of adapting user's work '''layand the (haracters.Learning - Agent has the ability of adjusting and"revising own behavior by utilizing the information ofthe external environment.Cooperation - Agents can communicate each other bysorne kind of Agent conununicatiqn language (forexample KQML or the FJPA-ACL language), and1l1uhtally cooperate to complete a big task.Mobility - Agents are able to move in, the network.These properties of Agent may be represented by thebasic model of Agimt system in Fig.1.
I1:1 i,I .,1 1, ~.,. I 1 II Q 1...1"1::; I'~1
i
",.'.
5. CONCLUSION
performance.
I
"". ""~ _ 00 _~ 00' ,,, ,,_. =,14/0ll/$2500rnJOS fEEE . 1
Yi Xiao, "Intelligent Infonnation Rel:Jieval Model Based all Multi-Agents'!]-
4244-1312-3/07/$25.00@2007IEEE. iPan Ying, Wang Tianjiang, Jiang Xtteli~g, "Bttilding Intelligeflt TI~Ii1mtionRebieva.lSystem Based on Ontology" 1-42#-1135-1/07/S251~1 @2iXYllEEE.Jae-Wc~) LEE, "A Model for Infom'!J.1.tion Rellieval Agent S)7stenl':skSed on
Keywords DiStribu~on" (}'~~Z177M9/07 $20.cx)(1d200~IEEE. .1"YanHU, Ymwan XUAN, Research on Web I.nfOli1k1.bOnExh'f1ch(nl ReSt-octonXNfL"97W-7695-33c'Ui/08$25.00@20081EEE. .1Qia.n 21m, Xianyi Q1eng, "The- Op:porhmities and Challenges of hlto11l1ationExh'action," 97R----fl-.7fi95-3305-0/08 S25.lYl @~ JEEE. '1
)fle-WOO l.EE1 "A Mode} for Informuoon Retdev<ll.Agent System &500 onI
Ke)'v.'oni':; Distrihution" 0-7695-2J17-9j07 $20.00 ([)200lIEEE. 1David He. Du, "lntelligent Storage for Information Rellicval" ~7M5-2452-
4/05$2iJ.OO@2005llir:E. I]lANG \',in-!-11l.'l.l, llU Yong-.Min. "A New Artilici;il. hltelligent htfo~natlon
Retriev.ll Methods" 978-0-769.J.-3559-3/09 $25.00@2OJ:),IEEE. l.R. Kalavathy*, RM. Suresht, R. Akhilat "Kdti ,md data ntin~ng".DRIDl-, "QltologyMB..'tsed hlfonn..ltiOll Retrieval: O1erview an:l. New
Proposiliml", ,I 'Panjce Panoy, Saiso Dj7.erOSki, Larisa N. SoI.datova, "OntoDM: An Ontology'of Data1v1ining" 978-0-7695-3S03-6j08'$25.00@2OC18IEEE. I .
Chen-Yu Le€' Vo.n-Wun 'S(X)'I_, "Ontology=based Infonnation Retrieval
andExtrilction" 0-7803-8932-8/05/$20.000 2003 IEEE. . I .Bill KlIlcs, Jack Kustano¥.itz and Ben SlUlLiderrnan, '''CategorizJlg .WebSearch Results into Meaningful and Stable Categories Using Ffu,,J~Feature
Techniques" ACM1-59593-334-9/06jlXOS ..:$5.00. ID Evangelin Geetha G Krishna Nilidu TV Suresh Kumar K T'\8jan~Kantll,
"S~m~lative PeIforn~<lllc:Evaltli.ltion of Inf0ln. mtion Retrieval Systel~ls" Cfl$-O.../695-3.l.36-6j~ 5:bJJOV 2fXlSIEEE I.Del Bimbn," A Perspective ViCl<\.'on Visua1l1ltormation Relrk'val Systems".[16]
[15]
[12]
[14]
[13]
[10][11]
[8J
[7]
[5]
[6]
[4]
[2]
[3J
E~i;:!-D1,,-".1
Fig, 3[4] A model for 'information retrieval agent system
The model is composed of nvo agents systems, the one isabout analyzing documents of server systemsarid the other isconsh'ucting profiles for documents' browsing. As theanalyzing documents agent, for key paragraph extraction weuse meaningful term's 'frequency and the kf'Y v'mrddistribution characteristics in a document, and those terms areselected by using,stemming, filtering stop-lists, synonym forsearch meaningful termS in a document. The agent receives aweb client's information retrieval request and extracts keyparagraph with fr'equency and distribution using thekeywords of the client. And constructing. profiles agentconstructs profile 'of the documents with the keywords,. keyparagraph, and address of .the document browsing. And thenmatiy clients can search many documents or knowledge.easilyusing the profile for information retrieval and browse thedocull'lent.
ABOUT THE AUTHOR:
-----_._-.Dr. R. P. Sillgh received B.Tecll. lInd Ph.D. degrers
ill electrim! engineering frOIll K.N.l.T; SlIltl7lijlllr
(U.P.) fwd lnstitllfe of Tedlllology, Ballams HindI!
lluit'CI"sity, l/1In1Jlnsi (U.P.), Illdill, ill 1~S81l1ld 1993,
l"('spccfh'Cly.Dr. Sillgh is II per5011 (~rllllllfidlJIICIl5ioJl(II pCI'sol/ality
witll 23 years afcxperience ill tile field of teachillg Ill/d
1'es('(lr(/, {Hellldill:? 4 .11m!':::lIS Direcft11j Prillcli1111. Dr. '""""-~-- ISingll is 11llU'JI//!a ~1val"i~ms flmdclJlic societies (!f 11l1tiollal fwd illtl'll/wtiQIIn:l
repl/te. He has fo IllS credlf 40 research pi7pers fllld nuthored follr boats. TlJE':~Cbooks are ,{lIell received /Jy tile studenfs &' awdemicial/s aUke. III 199.4-199p lie TOI15
tllC recipient or "TI'e Presidellt ofTni1in" as Direcfol! Principnl mvnrd fiJrr'esenrcll
contribufiol1s. C1Irrently/lE' is working as Dil'cctOi/ [Jl"llIdpn/.ilf Bllllia Dez,i
POOI/,WI Yndnu o/Jtnillcd B.Trell ill ComputerElIgg.&5ciorcc from. Kurnltsllt'tm University
KlIrukslletm, Indin null A1.Trcli ilt 11lfOl"llwtion
TccJmoTogy frOIll Gill"/{ G07.1illd Sil/gh lllrlrnpmstlm
Ulliver:::ity ill 2002 (Illd 2007 respectively. Slle is
mrrentljl wurkiJlg as Assistant Professur ill D.A.VCollege of fllgg. [I Teclillology, Knl/illn
(Mollindergnrll), PrCR!/ltlv slle is pursuing her PhD. . 'I'esenrcll at NJ.i\1S UniPCfsity, faiplll". Hcr researdl ----.."'1"' ".
interests ii/dude llljorJl/(/tioll Retrieval; Vveb bflsed retrievnl (Illd SewaHtic jlVcb
etc. Mrs. Yadav I::: nlife tillle member ojJlldin1l Society for Tec/mienl E!~licatio!1'
and Iwr email idis:pouHalll.il@glllai!.com. !
REFERENCES
[1] JING- YAN WANG, ZHEN ZHU, "Frame work of n~ulti-agcnt infOlmation
,!here are some significant attempt to .merge i~ormation.retrieval and ontological. :models. ,The integration of agenttechnology and ontology'. is a significant impact on theeffective use". of the web services. Macl:tine learningTechniques \viII be used for automatically extra~tingsemantic information .and text categorization. The evaluationmethodology based on using standard test collections assumesthat the.relevance can be approximated by ~ypical simiIal..ity ofthe query to document and hence user is not required to makerelevance assessment about retrieved documents.
The three .main challenges in utilizing ontologies inirHonnation retrieval are
]. To make sure that irrelevant concept ,,,,,ill not be kept,and that relevant concepts will not be discarded.. Descriptions of donihlents in infor"mation .retrievalare supposed to reflect the content of documents andestablish the foundation for the retrieval ofinformation when requested by users.
2. To map the information in documents and queriesinto.the ontologies.
3., To improve retrieval by llsing knowledge aboutreta.~iationsbetv.;een concepts in the ontologies.
.~: -I;.~"
Edlfcr.,fim/i1I Society's Gmf/p. of !lIst-itufio;ts (ll1te-g"mted Call/pus) jH KlI(rwledgel.ur/(, Faridl1[J,ld lind llis I'I1111if.ir{ is';' [email protected]'lJI
i
intelligent Information Retrieval in Data Minin~, , Poonam Yadav, Ravindra Pratap Singh ,I
Abstraqt: In this paper we prese"nt the methodologies and challenges of information retrieval. We will focus on data mining, datawarehousing. information- retrieval" data mining ontology, intelligent infon-nation retrieval. We ,will also focus on fundamental c6nceptd .behind all information retrieval methods. The Data Mining Ontology. is a .formal. description of the' domain concepts. and the relation. Qe.tWeel~the' concepts; it uses a hierarchical structure containing the concept entities 'and their'relation. We discuss some issues ,and challengeJfacing developing the new techniques targeted to resolve some of the problems associated with,intelligent information retrieval.
Keywords: intelligent Information retrieval, information extraction, ,data ,mining ontology;.---- ------1.0 DATA WAREHOUSING
Data warehousing is a process of centralized datamanagemt"'nt and. retrieval. Data wal"ehcusing r:~presei1l's ::mideal vision ,of 111aintai11ing a, central repository of all)rganizational data. Cenh'alization of data is' needed tomaximize user access and analxsis.as described in Qi Luo [1].A data warehouse provides the right .foundation for datam.ining. The data warehouse makes data mining more feasibleby removing many of the data .redundancy and systemn~anage111ent issues allowing users to focus on analysis.'
2.0 DATA MINING
Knmvledge Discovery and Data Mining are speedilydeveloping areas of research that are at the junction of severaldisciplines, including statistics, databases, Ai, visualization,and high-performance and' parallel computing. KnowledgeDiscovery and Data Mining goal is to' "turn data intoknmivledge." Basing on' it, Data mining is the core part of theKnowledge- Discovery' in Database (KDD) process shown inFig. 1 as described in Qi Luo []],
i1~,;~'.t4n~,
Data 'mining is the process of analyzing data from differentperspectives and summarizing it into useful information. Datarnining software is one of a number of analytical tools for
i,
i
analyzing data. It allm'lls users to analyze data 'froin ~nanrdifferent dimensions or angles, categorize it and si..llm~~a-r]zethe relationships identified.
111€KDD process may consist of the follmving steps: !"1 Data selection2 Data cleaning3 Data transformation4 .Pattern searchihg (data n~ining)
Data mining and KDD are often used interchangeablybecause data mining is the 'key to the KDD 'proc~s5 asdescribed in Qi Luo[I],
2,1 Data Mining TasksDifferent methods and techniques are needed to find dif eT€ntkinds of patterns. Based on the patterns tasks in data nfiningcan be classified into Summarization, classifiCf\tion, dU,5,,' ring.association and h'end Analysis as described in Qi Lua [1]
1. S1.1m11lC'lTizanon.Summarization is the abstract~on, 0r" '
generalization of data. A set of task-relevant qata issummarized and absb'acted. 1l1is results in a s+aller 'set 1Nhich gives a general overvievF of the i data,usually with aggregate information. !
2. Classification. Classification derives a functiJn ormodel which determines the class of an object l,asedon its attributes. ,I
3. Clustering. Clustering. aims to establish relativl~J.yhomogeneous subgroups In the gIven data. I
4. Trend analysis. Time senes data are l'icordsaccumulated over time. Such data can be VIeW[d asobjects WIth an attribute time.
2,2 Data Mining TechnologyData mlllll1g adopted its techmques from l)'c\I1Ylesearch Mee\S
including statistics, machine learning! associatiQn Rules,neural networks as described in Qi Luo (1.]. 1_
CJ) Association Rules. Association I'u'le generators' hTe apowerful data mining teclu\iqul2 used to search thtOl.lghan entire data set for rules revealing the naturJ andfrequency of relationships or associations betwe€l~ dataentities. TIle resulting associations can be used to! filterthe information for human analysis and possib1y todefine a prediction model based 0;1 observed behaVir)f,
,'! ,~., -",;,' .•• ".
Oat'a type
~'Iethod
rU!1cLivn
"- ... -.'''''.''-.-.-. __ .''
Data .11.1jningpn,'lcess
For a selected data, firstly, pre-processing is preformkd . 011I
the data, such as filling the missing values, cleaning the data,or exh'acting features £1'omthe data. Then one of the in<.{uctionalgoritluns is introduced for mining the pre-processetI dataac~o~ding .to the task, for example, if.the use~s ~'ant to fifld ~herelationshIp beh-veen-temporal data Items wlthm a h'anSClction
I
database, associCltion rule mining can be used. After that,hypothesis testing may be applied to check the im~ortantrules and the mined rules are given to the us~r for
I
I
I
I
I
concerns \'vith Cllgorithms intro(h](eL~bv Favyact _shown in-Fig:3. . .. I
Some algorithms can only be used lor some specific Idf.m ofdata type. Thus each algorithm should be within the sub-treeof data .type. For classify the structure, we arrange I eachalgorithm in some specific function. Note some algorithmsserve for multiple functions, for example, SVl\1 can bel usedbotl, for classification and regression. In. this case, thisalgorithm will appear in each f~nction category.
Fig. 3: The data mining proqess
3.1 The Structure of Data Mining Ontology
interpretation.
Fig. 2 The Problem solving method Heuristic classification
(2) Artificial Neural Networks are recognized in theautomatic learning fr~U11evvork as universalC1pproximations, vvith massively parallel com.putingcharacter and good generalization capabilities, btit alsoas black boxes'due to the difficulty to obtain insight intothe relationship learned.
(3) Statistical Techniques. These include linear regression,discriminate analysis or-statistical summarization.
(4) .Machine learning (ML) is the center of the data.mini';g:concept, due to its capability to gain physical insight intoa problem, and participates directly in datu selection and.model sear~h steps.
2.3 Heuristic Classification in Data MiningClancy & Fayyad etal[2]. Proposed a general problem solvingmethod called Heuristic Classification shown in Fig 2. Thedata mining is a goal-directed task. There are always l1lCldelsand, patterns within the data. ,Consh'uding a successful datanlining system needs to k~ow' what to look for in prior._Theusers must provide a wen-defined problem that can be solvedby' the datCl1l1iningC1lgorithms.
3.0 THE DATA MINING ONTOLOGY
By the heuristic c1qssification for data .mining, the userobserves the data and algoritluns locate in the -right side of Fig.2. the heuristic. n\atch -pr~cess matches a set of algorithmssati~fying the proposed goal As one algorithn1,alone normallycCln't solve_ the task, the algorithms lilay also need to besubdivided into sub-procures described by t~e refine step inFig. Tha whole datClmining project can be identified by thehe-urjstic c1assificCltionmodel with the following steps:
• Ide'ntify the mining task.• Match the task to ohe sol~tion within a solution set.• Decompose the soluj:ion to several sub-algorithms.
Ontology is a formal description of the domain concepts andthe relation between the concepts; it 'llSe,Sa hierarchicalstructure containing the concept entities and their relation.Ontology "iV.ouldbe Uliderstandable for both human beings'and machine and could :LJereused.' For constructing ontol'ogyfor datCl mining, we need a model for mode~ing the datCl:rnining algorithms. ,"Vetake the data Inining modeling only
4.0 INFORMATION RETRIEVAL (IR) .~
In IE, we retrieve only structured data. And lR provides alltypes of informa tion access such as analysis, organiZe tiOl1,storage, searching and retrieval of information. As aceol'dingto SlatOl~'Sclassic textbook: "Tnformation retrieval i~~ll-fj(j'ldconcerned-with the struchned, analysis. orgClllization,stnrClge,seClI'Ching,and retrieval of information" [51, I
I
I
I
6.0 ONTOLOGY AND INFORMATION RETRIEVAL
Information retrieval is defined by Carlo Meghini et a1."lnformation Retrieval as the task of identifying docurn.ents ina collection on the basis of properties describe.d to thedocuments by the user requesting the retrieval" [4].
Combination of IR and database will be valuable for thedevelopment of probabilistic models for integrityunslructurect semi-structured and structured data, for thedesigl~ of effective dishibuted, heterogeneous informationsystems [5]. Jimmy Un describes the information retrievalcycle as given in figure4 .
The Information Retrieval Cycle
Figure 4: Information Retrieval Cycle
. In. figure4, source selection is first step in. IR cycle.Appropriate resource is selected for all valuable data. Thenquery will' be formed according to user needs in queryL1flllulation. Query will be processed for searching desired
results, rank list will be prepared -according to .search resultsaI.ld appropriate documents are selected.' If ~election is notaccording. to user needs, 'selection criteria goes back to sourceselection and query. formul"tion. Examinatioll process ch~('ks
.the, documents for,validation, if its validation is pr~ved' thendocuments will be the result otherwise output of examinationgoes back to source reselection.and query formulation.
Various challenges in IR:New challenges to IR community and mo~vated researchersto look for intelligentInformation Retrieval .(IR) systems thatsearch and/or filter information automatically based .on some~ugher level of understanding are required. \A/hat type ofsemantics is to be .used to improve ef~ectiveness in lR? \-Vhatwill be th.€ hypothesis to improve representation of documents
. and by incorporating limited semantic knmvledge? [7].
5.0 INTELLIGENT INFORMATION RETRIEVAL SYSTEMS
JIANG Xinhua et 1"11 described thiit the retrieved. documentshave the right terms but may not be in the desired cont~xt. Ithas been argJect that exploiting the .lIser's context h~s thepotenti,11to improve the performance of information .ret~'ievalsystem described in [10].
111e challenge of developing intelligent infornlationI
retrieval.system based on context of dotuments: '• How to develop a'lnethodology of assigning context
to documents using assigners and validate the resultsthrough n{easurement of consistency be~eenassigners?
• Web usage tools, usually implementing the process ofcustomizing the content and the structure of "\-'.'eb sitesin ordpr to ~atisfy the specific need of each and ~veryuser, without asking for it explicitly;
• Web content mining tools used for automaticcJ<1ssifjcationl)f dOCllment contents, including itheirmultimedia obre-ctsas \vell as textual information.
!
JING-YAN WANG et a1. described that the characteristics ofan ideal information reb.ieval system are that searching is Heetand result is accurate described in [9]: I
,
Ontology is the formalized description of share con(e~')ttlalmodel to the information resource, so any conceptual object Ccan be defined as: C = {D, '''l, R} Where,. D indicates suchdomain knowledge, VVis a status set of correlative thitlg inapplied domain; R is a set of concept relation'in domain space{D,W}.
It has two effects that ontology is used in informationretrieval. I
1. Automatic anal\'sis to the domain attribute ofdocument.The in.formation retrieval system Jooks for ke)',,~'ords
, I
acc.:'1rdingto a certain means in seal'.::heciinfonn~tior"dO(lJmcnt; these keywords can be used in documentclassification which is dependent on ontq>logyknowledge. So information rehieval, system cmi sort,documents to some possible domains combi\1ing' withprimary content of information document, thenfiltrates off irrelevant domains, gets final resllIt ofcorrelative domains which the document is classed to.
2. Intelligent formulation and visualization to user'ssearching demand.
It is very important that 'how to visualize user's know~edgedemand by appropriate Hlethod in. preliminary stage ofinfonnation retrieval.
. >."
ABOUT THE AUTHOR:
answering", 2))4.O,en-Yu Lee ,Von-Wun seD, " Ontology based Jnformation'Retri~v'iil arld
Extraction" IEEE 0-78ll.'.-8932-8/OS 200..1:). i
Hao llatl and Takchiro Tokuda .' "A Method [or integ .•..ationj ofvVebApplication::; Based on lnformption EXh'action" ICW"E.2008. IEEE 00110:J109/lC
Seh.'Etian A, Ryos-, Juan D. Vel'asqttC'zt, Eduardo S, Vi.?rats, HiroshIYasudil_ dnd Terumas<'! /" Improving Web Site Content Using a Goncept-ba'Jeti ,Knowledge Di';covely" JEEE/WJC/ ACM hltcmatimlal CoAferenccon Web lntellige:nce2()(J(" I . .
GillC'S Nadl0u.ki, "A ~v1elhxi for Information Exl:,raction.fr01l1 the ~Veb"o-'
7S03-9521-2/0Ii/52o.m)2OOI IEEE.Hao Han and Takehiro Tokuda, "A Method fOr Integration ,oMeb
Applkations Bas.ed onInfommtion Extraction".Xmnming liu , Weihua li , Shixian Li3 Zhenwen Tao, "A New Model toDescribe Information Setnantic" Intem •.ltional Symposium on hlfohnation
Science and FJlgi<''eru1s.2.Xl8 IF.cE . I
Yi Xiao, Ming Xiao, Fan Zh.,mg. "Intelligent Information Reh;eval ModelBased on Multi-Agents" 14244-1312-5j07/S25.00@2fiJ7IEEE .1 ...:.;Mohan S. Kankanhalli and Yang Rui, 'i ApplicatiOll Potential of M.~ti1UedialJ1fOlmation Rcbi.eval" IEFE I VoL 96,No. 4, April200S'OOl&-9219/SEfl(j ,
p~~~::",mgTCu~iang. JCTI'g Xuc1ln& "Building h,telligent h"lfoLalionRetlieval System &lsed on Ontology" "1-4244-'1133-1j07/$25.oo (~71EEE
[20J
[19]
[17]
[16]
[15)
[14)
POOI/(1I1I Yadl1v obit/iI/cd B.Treli ill Compllff'r
El/gg.c'T5cicnce frow Kllrukslletm Ulliversi~lf
Kllfllkslletm, India alld M,Teell ill ll1jonllatioll
Teclmology frolll Gllru GOllind Sillg;' TndmpmslJm
U/liFt'rsily ill 2UU2 and 2007 respecfivdy .. Slie isclIrJ'eI1t1y' 'lporkillg 115A~sisfl1l1t Professor 111 D.A.V
College of Ellgg. & Technology, Knuinn(Mdlrilldergl1rld. Presently she is pursuing Tier PhD.reSCl1rc11at NIMS L/llitic/'sitlf, laipl//'. Hl:'r research ."'1""lIltere~ts lIlclllde {ufO/llIatlOll Retlleol1l; Well !Jased ,etnef'nllllld Sell/llnfJc tVi'll
etc MIS Yadav IS Illife tlllle 1I1cml1cr oj I/ldlan SOCiety jm Tecllll/cllT Erl.rllaflOlI
and III'/' ell/llI/,d 1<; POOIIl1I1/lI@gllllll! C01ll
01". K p, 5111gl1recemed s.rIccl,. /Jlld PI/D. degrees ---- -
ill electrical cngillecrilJg froll! K.N.J.T; 511ffn./lpllr
(UP.) and Institl/te of Technology, [kmnms Hindu
Uni7.1ersify, \/amilllE'i (U.P.), rlldin, ill 1~188 11.t1d.i99:',
l'e~:p';cti{!ely. !d :Dr. Sil/Slt I::; II pnS{)/l 0/ !lllIltidhl/l~l/::;iOIl<11 f1(,/"~0I1t1!i111 ;
witl! 23 y.cal's ojc.tpcl'icllce il! fllefidd OjtCllcl:illg(lll~1 !'. ' ~research 11Ie/l/{llIIg 4 years IlS DJreetOlj PrIIlctplll. Dr. - ---- -
5ingll is (j IIIcmbel: of variolls academic societies (>/ Illlfiollni nlId i,l fen ntiemal. '., . '.1repllit'. He has to his credit 40 researcll pnpcrs ,wd Ill/tllO.'"edfollr.boo~s.~liese
baoks an- well received hl/ tile st/lawts & acadelJlicinus nlike. 111 ]994-1995 'lie
wns t1le redrient of "T"~President ofJlldin" (IS.Directo'rl prillciptii mJ(lid for
research cO;ltriblltio!ls. CI/rrwtly he is worki!lg as Directol! Principl11, fl{ Bill/ll1
Oev! Edllcatiol1l11 Society's Group of IllstiflltiO!1S (Integrated Cnll/pus) JBKllowledgt: Park, Fl1ridIlIJ/ui'l1Il/{his email id is: I'[email protected].
[12]
[13J
[18]
Data \.\ral'ehousing is a process of centrAlized dataI1\allagement ilJld retrieval, Data mining is the process ofanalyzing,data from different-perspectives and summarizing it
into useful infOl'mation. It allows users to analyze dilta frommany different dimensions or ang'les, categorize it, andsummarize the relationships identif~ed Th~ Data MiningOntology is a formal description of the dbnfain COilCepts andthe relation between the' concepts, it uses a hierarchicalsh'ucture containing the concept entities and their rel~tion.Ontology would be 'understandable for both human beingsand machine and could-be reused.Information retrieval provides all types of information
access. Concept-based visual indexing has high expressivepower, which can easily. communicate with user but still;JwoJves information Joss in transforming vlsual111aterials intoiext, and requires more intensive human labor. InformationExtm'ction is a, logical step to retrieve structured data and theexh'acted information. The extracted information is 'then usedin searching, browsing, querying, and mining. TI,E' retrieveddocuments have the right terms but rnay not be in the desiredcontext. It has been argued that exploiting the user's contexthas the potentia I to improve the performance of informationretrieval systems.There are .two important issues to be addressed. First one is
t~ design and integrate an efficient ontology and the second isto establish of linkages, betweei, ontologies of differentdomains,
REFERENCES:
[1] Qi LtIO, " Advmring Knowledge Discoye:ty ~ Data Mining", 2IXl8 IEEE
001101109 /WKDD.2008.153 .[2] .NL:1.o-50ng Lin, Hui ZL1a.ng. ~d Zl"kmg-Guo 'Yu, "An OiltOlogy far
Sup~X)rtil~g Data Mlllil""tgProces..« M'l1lUScri:pt rcccivcdJune 30, 21X16.[3] . AnHili .Do<u, RRghu Rumkrishm, and Shiva Kumar, "Managing.
, Infol1l\ation EXtraction", STGMOD2006, AOv11-59593-256-9/QS[4) CARW MEGHlNT, FABRlZlO SEBAS11ANT, AND UMBERTO
.$TRACClA "A Model of :rvlu1l:imcrua h'l£onn'-l.tion Retrieval", Journal ofthe ACvl, V(,I 4R, No. 5,Scpt~l11Ll('r 2((11, pp. 9."}-)-970.
[5J G. Sc,lton,." Automatic TnfomlCltion Orgil11ization and RetJiev8.I", McGraw-. Hill,. N01'\' York,. 19(-,8.
[6] Qmllenges In Infol1l1ation Retrieval .Uld,l,nguage Modeling, Report of aWorkshop held at the center for in:telligent Infoinlation Retri.eval, Universityof Ma'i.o:;achuse~ Amhcrsl'and September 2002. '
[7] Jinuny Lin, "An inb'oduc6('1ll to infornwtion retrieval artd ,question
answering",2004.[81 Tanveer J Siddiqui "Intelligent Techniques for Effective hlfOlTIlation
Retrieval",2006.[9] jlNG-YAN WANG, ZHEI'J ZHU, "FRAMEWORK OF MULTI-AGENT
JNFORMATION REIRlEVAL SYSTEM BASED oN Ol'-'TOLOGY ANDITS APPLlCATION" 978-1-4244-2Jl96.4/08/$25.00@20081EEE
[10] JIANG Xin-Hual, LTU Yong:"Min, "A New ArtifiCial Intelligent h1fonnatioilRetJievaJ Methods" 978-0-7~3559-3/09 $25.oo@2C09IEEE
[11} JinmlY Lin, "An intrcduc~on to infonnation retrieval and question
7.0 CONCLUSION
Augmented Systems in Ontology BasedIntelligent Information Retrieval
Ravindra Pratap Singh, Poonam YadavAbstract: In order to solve the problem of information overload, current information retrieval systems need to be improved. Intelligenceshould be embedded to search systems to manage effectively search, retrieval, Filtering and presenting relevant information. This can bedone by ontology based information retrieval techniques. Systems are evaluated by using their information retrieval features. Ontologiesare introduced to describe the semantic information of knowledge leases in order to meet requirement of utmost sharing and reuse of thedomain knowledge .The paper provides an overview of ontology-based information retrieval techniques and software tools currentlyavailable as prototypes or commercial products,
Keywords: Ontologies, Information retrieval, Semantic indexing, Semantic mapping ..----------1. ONTOLOGY AND INFORMATION RETRIEVAL
JING-YAN et al. described that characteristics of an idealinformation retrieval systems are that searching is fleet andresult is accurate. Traditional methods of information retrieval'can be divided into three essential kinds: text retrieval, dataretrieval and topic retrieval as described in [1].Text retrieval system: The user's information retrieval
demand is represented as keywords which are compared witheach word of whole text in text retrieval system. Its searchingmethod is based on word-frequency analysis, and typicalsystem is Google. The excellences of this kind of informationretrieval system are includes a large amount of output resultsand super automatic search control. Its shortcomings are thatreturn information is accumulated and reality is too low.Data retrieval system aims at the structural information
system, retrieval demand and information data all stands bycertain formats, and is provided with the decided structure. Itpermits appointed field search, and various kinds ofcommercial service database all adopt this searching mode.The quality of data retrieval depends on code organization,and may cost a lot of money. The output result is relativeaccurate, but can not generate all expectancy information.
Topic retrieval system collects information by manpowermanner or semi-automatic manner. It scans accessingdocument by dedicated method, and append to somebeforehand determinate topic. User can visit downwardsgradually from one layer to next layer, so finally can getretrieval result. The representative system is Yahoo.The excellence of topic retrieval system is that user can
easily master searching process. Ontology is the formalizeddescription of share conceptual model to the informationresource, so any conceptual object C can be defined as:C~{D, W,R}Where, D indicates such domain knowledge, W is a status
set of correlative thing in applied domain, R is a set of conceptrelation in domain space {D,WI.
lt has two effects that ontology is used in informationretrieval.(1) Automatic analysis to the domain attribute of document.The information retrieval system looks for keywordsaccording to a certain means in searched information
document; these keywords can be used in documentclassification which is dependent on ontology knowledge. Soinformation retrieval system can sort documents to somepossible domains combining with primary content ofinformation document, then filtrates off irrelevant domains,gets final result of correlative domains which the document isclassed to.Moreover, it can be make keyword matching based on the
approximate semantic network of ontology, automaticallyclass correlative documents to a certain domain. In this way,information retrieval system searches some selectivedocuments which are correlative to user searching demand,and do no longer scan all information documents. Sosearching time and cost can be compressed greatly and user'ssemantic reiativity judgment may be expedited.(2) Intelligent formulation and visualization to user'ssearching demand.For the searching keywords given by user, we can effectivelyestimate feasible domain depending on ontology knowledge.Then information retrieval system enumerate consistentconcept and definition of the domain to user, user can makejudgment and selection. On the one hand, it can help userconfirm their knowledge demand, user can more distinctlyvisualize information demand which is not clearly presentedor little does user think. On the other hand, it can letinformation retrieval system ascertain the position ofsearching keyword in ontology, so help machine tounderstand user's searching intention, and provide moreaccurate and more relevant information.
Because of ontology defines essential kennel concept andtheir semantic relation, it can be used as the tools of resourcedescripnon and knowledge presentation.
2. INTELLIGENT AGENT TECHNOLOGY
At present, Agent widely has applied in the fields ofinformation retrieval and service, electronic commerce,distributed computing and so on. Yi Xiao et al. describedAgent is a concept developed in the field of the artificialintelligence (AI), is a hot spot in Al the recent years. Generally
3. CG INTElliGENT INFORMATION RETRIEVAL SYSTEMARCHITECTURE
Fig.1 [2]: The basic model of Agent system
The basic model of Agent system generally, there are threemost basic properties of Agent such as the autonomy, learningand the cooperation. A basic Agent system should has one ofthese properties, a advanced Agent system must have theautonomy and one of other two items, but an ideal Agentsystem must realize all of these properties and introduce othercharacteristics (for example mobility according to the actualdemand, etc).
Fig 2[3]: CG Information Retrieval System architecture
On the client side, users input queries in the retrievalinterface. On the server side, retrieval application accepts therequest of users, and then invokes Jena API to analysis CGontology and Forms search results, the final results arereturned to the users at last. I
4. INFORMATION RETRIEVAL AGENT SYSTEM BASED ONKEYWORDS DISTRIBUTION :
I
Jae-Woo LEE described that for efficient information r~trievatit is important that keywords are defined very ~ell asappropriate terms about all documents in the IntJrnet ordistributed computing systems. And there are three kinds ofalgorithms, retrieval algorithms, filtering algorithm andindexing algorithms. Retrieval algorithms are extractingimportant or meaningful information from database,knowledgebase or documents. Those are classified tosequential scanning and searching indexed text files. Filteringalgorithms are simplifying result of text or information forefficient information retrieval. Indexing algorithms areconstructing data structure for searching information exactly.Indexing is one of the most important data for efficientinformation retrieval. For automated indexing there arevarious algorithms, stemming, stop list, thesaurus, etc asdiscussed in [4].
Stemming is the process of matching morphological term'variants for using general terms. Stemming is for reducing thesize if index files and improving information retrieval system's
Inquiry system is B/S model, Java program which invokeslena API functions to realize the complex Semantic inquiries.Jena is a Java framework for building Semantic Webapplications. It provides a programmatic environment for RDF(Resource Description Framework), RDFS (RDF Schema) andOWL, SPARQL (Simple Protocol and RDF Query Language),and includes a rule-based inference engine. Jena can be used tocreate and manipulate RDF graphs. The system mainlyincludes: the user retrieval interface, the retrieval application,Jena Ontology analytical tools, and the CG ontology. TheSystem architecture is shown in fig.2.
[. . __.:--_'."JU~'Crr-cfn-c'.n:ai lnterJ~ec'----1 ---..---.- --..
C\!icnl side U~e t'~"'quest .
~ :V-:'-':~I:-_ ..- ~ - - -ff"LtCr ,"vii tre"" ",,----[~l~=~~~~,:,-~;~~~:~lit.J
;1, .'<-'\nall_)"1_ic~..t rc:s,u1ts-L~:~ta(}rnolog.v ,AP~
R<,,~--'---- ----.. ;1 R (;~Ifurn
C"......•......................]C<-; 'lultology
•..._----,- ._---_.,
Users input keywords which they want to search, thenretrieval system return the matching document to users.Because of synonyms and polysemy, it's very difficult tounderstand user exact needs by keyword. Often the initialkeywords do not get the resulls they want, the mildly relevantor irrelevant documents were also retrieved. Pan Ying et aldescribed that CG introduces how to use the computer togenerate, process and display graphics. CG is very broad in itscontent. lt is difficult for learners to understand the wholesystem of disciplines as described in [3].
believed that, Agent is the intelligent entity which can runindependently and continuously under the environment ofisomerism computation, and has the properties as follows asdescribed in [2].
1. Autonomy - Agent is able to run continuouslywithout the disturbance from outside or nobody, andcan control its own behavior and internal condition.
2. Adaptability Agent can sense the externalenvironment, and carries on the self- adjustmentaccording to the change of external environment,therefore has the ability of adapting user's work wayand the characters.
3. Learning' Agent has the ability of adjusting andrevising own behavior by utilizing the information ofthe external environment.
4. Cooperation - Agents can communicate each other bysome kind of Agent communication language (forexample KQML or the FlPA.ACL language), andmutually cooperate to complete a big task.
5. Mobility. Agents are able to move in the network.These properties of Agent may be represented by thebasic model of A~ent system in FiKl.
ABOUT THE AUTHOR:
performance.
--~:::;;-l
--I"" ..",..•.j
Fig. 3[4] A model for information retrieval agent system
The model is composed of two agents systems, the one isabout analyzing documents of server systems and the other isconstructing profiles for documents browsing. As theanalyzing documents agent, for key paragraph extraction weuse meaningful term's frequency and the key worddistribution characteristics in a document, and those terms areselected by using stemming,. filtering stop-lists, synonym forsearch meaningful terms in a document. The agent receives aweb client's information retrieval request and extracts keyparagraph with frequency and distribution using thekeywords of the client. And constructing profiles agentconstructs profile of the documents with the keywords, keyparagraph, and address of the document browsing. And thenmany clients can search many documents or knowledge easilyusing the profile for information retrieval and browse thedocument.
5. CONCLUSION
There are some significant attempt to merge informationretrieval and ontological models. The integration of agenttechnology and ontology is a significant impact on theeffective use of the web services. Machine learningTechniques will be used for automatically extractingsemantic information and text categorization. The evaluationmethodology based on using standard test collections assumesthat the relevance can be approximated by typical similarity ofthe query to document and hence user is not required to makerelevance assessment about retrieved documents.
The three main challenges in utilizing ontologies ininformation retrieval are
1. To make sure that irrelevant concept will not be keptand that relevant concepts will not be discarded.Descriptions of documents in information retrievalare supposed to reflect the content of documents andestablish the foundation for the retrieval ofinformation when requested by users.
2. To map the information in documents and queriesinto the ontologies.
3. To improve retrieval by using knowledge aboutretaliations between concepts in the ontologies.
REFERENCES
[1] JING-YAN WANG, ZHEN ZHU, "Frame work of muIti-agent infonnation
retrieval ~ystem based on ontology and its application" 978-14244-~
4/08/$25.oo@2008IEEE.[2] Yi Xiao, "Intelligent Information Retrieval Model Based on Multi-Agents" 1-
4244-1312-5/lll /$25.oo@2OO7IEEE.[3J Pan Ying, Wang Tianjiang, Jiang Xueling, "Building Intelligent Information
Retrieval System Based on Ontology" 1-4244-1135-1/lll /$25.oo@2007IEEE.[4] Jae-Woo lEE, "A Model for lnfomlation Retrieval Agent System Based on
Keywords Distributiun" Cl-7695-Z777-9/lll $21]'00@2OO7IEEE.[5] Yan HU, Yanyan XUAN , "Resean:h on Web Information Extraction Based
on XML" 'l7&{)-7695-2..TI4<i/08$25.oo@2OO8IEEE.[6J QianZhu, Xianyi Oleng, "The Opportunities and O1aI1enges of Information
Extraction" 'l7&{)-7695-35(Jj..()/08$25.oo@2OO8IEEE.[7] Jae-Woo LEE, "A Model for Information Retrieval Agent System Based on
Keywords Distribution" D-7695-Z777-9/lll $21]'00@2OO7IEEE[8] David H.e Ou, "Intelligent Storage for Information Retrieval" 0-7695-2452-
4/05 $2D.oo@2005IEEE.[9} JIANG Xin-Hua1, UU Yong-Min,. "A New Artifidal Intelligent Wonnation
Retrieval Methods" 'l7&{)-7695-3559-3/09 $25.oo@2OO9IEEE.[10J R Kalavathy', RM. Suresht, R Akhil4, "Kdd and data mining".[11] DRIDI-, "Ontology-Based Infommtion Retrieval: Overview and New
Proposition" .[12] Panjce Panov,Sajso Dizeroski, Larisa N. Soldatova, "OntoDM An Ontology
of Data Mining" 'l7&{)-769s.:35ffi.6/08 $25.oo@2OO8IEEE.[13J Olen-Yu Lee' Von-Woo 500'-, "Ontology=based Information Retrieval
and Extraction" D-78tM932-8/05/$20.oo 0 2005 IEEE.[14] Bill Kules, Jack Kustanowitz and Ben Shneiderman,. "Categorizing Web
Search Results into Meaningful and Stable Categories Using Fast-Feature
Techniques" ACM 1-59593-354-9/06/0006 ...$5.00.[15] 0 Evangelin Geetha G Krishna Naidu T V Suresh Kumar K Rajani Kanth,..
"Simulative Performance Evaluation of Infonnation Retrieval Systems" 978-
D-7695-313(;<;/08 $25.oo@2OO8IEEE.[16] Del Bimoo," A Perspective View on Visual Information Retrieval Systems".
Poonnlll Yfldav obtained B.Teel, in ComplIterEngg,&Science frolll KI/Yllkshetm UniversityKliruks/tetra, India and M.Tecfl in In/orlllationTeelmvlogy frolll Gum Govind Sing!' lndrapmstlmUniversity in 2002 and 2007 respectively. Sl,e is
currently working as Assistant Professor in D.A.VCollegr of Engg. & Technology, Kanina(Mollindergarll). Presently slle is pursuing her P1I.D.research at NlMS University, jaip"r. Her researcllillterrsts inelllde Information Retrieval; Web based retrieval and Semantic Webetc. Mrs. Yadav is a life time mell/ber of Indian 50dety for Technical Educationand her elllai! id is: [email protected].
Dr. R. P. Singh received B.Tecll. and Ph.D. degreesin electrical wgineering from K.N.J.T; SlIltanpur(U.P.) ml/t Institllte of Technology, Banams HinduUnivrrsity,vamnasi (U.P.), Irldia, ill 1988 and 1993,resprclivcly.Dr. Singh is a person of lIIultidimensiollal personalitywitl, 23 years afexperience in tile field of teaching andresearell incillding 4 years as Direclor/ Principal. Dr.Singll is a member of various academic societies of national and internationalrepute. He has to /lis credit 40 researell papers and alltllored four books.Tllesebooks are well received by tT,estudents & academicians alike. ill 1994-1995 Ill'wastire recipient of "Tile President of india" as Director/ Principal award for researellcontributions. Currently lie is working as Director/ Principal ill Binlla Devi
Educational Society's Group of Institutions (lntegmted (nIl/PIlS) JB KnowledgePark, Faridnbadand his email idis:[email protected]
!
Intelligent Inform'ation Retrieval in Data Mining. I
.Poonam Yadav. Ravindra Pratap SinghAbstract: In this paper we present the methodologies and challenges of information retrieval. We will focus on data mining, datawarehousing, information retrieval, data mining ontology, intelligent information retrieval. We will also focus on fundamental conceptsbehind all information retrieval methods. The Data Mining Ontology is a formal description of the domain concepts and the relation betweenthe concepts; it uses a hierarchical structure containing the concept entities and their relation. We discuss some issues and challengesfacing developing the new techniques targeted to resolve some of the problems associated with intelligent information retrieval.
Keywords: intelligent Information retrieval, information extraction, data mining ontology;.----------
1.0 DATA WAREHOUSING
Data warehousing is a process of centralized datamanagement and retrieval. Data warehousing represents anideal vision of maintaining a central repository of allorganizational data. Centralization of data is needed tomaximize user access and analysis as described in Qi Luo [1].A data warehouse provides the right foundation for datamining. The data warehouse makes data mining more feasibleby removing many of the data redundancy and systemmanagement issues allowing users to focus on analysis.
2.0 DATA MINING
Knowledge Discovery and Data Mining are speedilydeveloping areas of research that are at the junction of severaldisciplines, including statistics, databases, AI, visualization,and high-performance and parallel computing. KnowledgeDiscovery and Data Mining goal is to "turn data intoknowledge." Basing on it, Data mining is the core part of theKnowledge Discovery in Database (KDD) process shown inFig. 1 as described in Qi Luo [1].
Fig. I Knowledge Discovery in Dlltaha'iC (KDD:f pro('C'.~
Data mining is th~ process of analyzing data from differentperspectives and summarizing it into useful information. Datamining software is one of a number of analytical tools foranalyzing data. It allows users to analyze data from many
different dimensions or angles, categorize it, and summarizethe relationships identified. i
The KDD process may consist of the following steps':1 Data selection2 Data cleaning3 Data transformation4 Pattern searching (data mining)
Data mining and KDD are often used interchangeablybecause data mining is the key to the KDD prdcess asdescribed in Qi Luo[I]. j2.1 Data Mining TasksDifferent methods and techniques are needed to find ,ifferentkinds of patterns. Based on the patterns tasks in data' miningcan be classified into Summarization, classification, clJstering,
I
association and trend Analysis as described in Qi Luo [~]1. Summarization. Summarization is the abstrattion or
generalization of data. A set of task-relevan~ data isI
summarized and abstracted. This results in a!smallerset which gives a general overview of tt~e data,usually with aggregate information. i
2. Classification. Classification derives a function ormodel which determines the class of.an objebt basedon its attributes. I
3. Clustering. Clustering aims to establish r~lativelyhomogeneous subgroups in the given data. I
4. Trend analysis. Time series data are I recordsaccumulated over time. Such data can be viewed asobjects with an attribute time. !
2.2 Data Mining TechnologyData mining adopted i.tstechniques from many research areasincluding statistics, machine learning. association:. Rules,neural networks as described in Qi Luo [1]. I
(1) Association Rules. Association rule generators are apowerful data mining technique used to search ~hroughan entire data set for rules revealing the nat~lTeandfrequency of relationships or associations betwJen dataentities. The resulting associations can be used "to filterthe information for human analysis and pos~ibly todefine a prediction model based on observed behkvior.
(2) Artificial Neural Networks are recognized I in theautomatic learning framework as universalapproximations, with massively parallel co~puting
3.1 The Structure of Data Mining Ontology
....... _,-,.,--".,.Data ruining process
- .~lo:ll'" ?r"I~oce~mg.1 P''"l,oCClSoi!
dll' Tl'1!" "l dtI.J)lI'-
I
P""f",,1ilij,'S Inriu,lkll1
koowledge im("lpt«~liot:l \ AI~l)f'ithmMOil.'F~lfml~
IIFig. 3: The data mining process
For a selected data, firstly, pre-processing is preformed on,
the data, such as filling the missing values, cleaning the data,or extracting features from the data. Then one of the inductionalgorithms is introduced for mining the pre-process~d dataaccording to the task, for example, if the users want to ~ind therelationship benveen temporal data items within a transactiondatabase, association rule mining can be used. Aft~r that,hypothesis testing may be applied to check the importantrules and the mined rules are given to the u~er forinterpretation.
character and good generalization capabilities, but alsoas black boxes due to the difficulty to obtain insight intothe relationship learned.
(3) Statistical Techniques. These include linear regression,discriminate analysis or statistical summarization.
(4) Machine learning (ML) is the center of the data miningconcept, due to its capability to gain physical insight intoa problem, and participates directly in data selection andmodel search steps.
2.3 Heuristic Classification in Data MiningClancy & Fayyad et al[2]. Proposed a general problem solvingmethod called Heuristic Classification shown in Fig 2. Thedata mining is a goal-directed task. There are always modelsand patterns within the data. Constructing a successful datamining system needs to know what to look for in prior. Theusers must provide a well-defined problem that can be solvedby the data mining algorithms.
,
Some algorithms can only be used for some specific form ofdata type. Thus each algorithm should be within the sub-treeof data type. For classify the structure, we arrange eachalgorithm in some specific function. Note some alg6;ithmsserve for multiple functions, for example, SVM can be usedboth for classification and regression. In this case, this
h f ,ialgorithm will appear in eac unction category.
Fig. 2 The Problem solving method Heuristic classification
By the heuristic classification for data mining, the userobserves the data and algorithms locate in the right side of Fig.2. The heuristic match process matches a set of algorithmssatisfying the proposed goal. As one algorithm alone normallycan!t solve the task, the algorithms may also need to besubdivided into sub-procures described by the refine step inFig. The whole data mining project can be identified by theheuristic classification model with the following steps:
• Identify the-mining task.• Match the task to one solution within a solution set.• Decompose the solution to several sub-algorithms.
1:)3ta t)t'Pc
Function
Method
AIg oritll n 1
C':on;:.;;tndnt. po cr rornuJJ] cc
3.0 THE DATA MINING ONTOLOGY
Ontology is a formal description of the domain concepts andthe relation between the concepts; it uses a hierarchicalstructure containing the concept entities and their relation.Ontology would be understandable for both human beingsand machine and could be reused. For constructing ontologyfor data mining, we need a model for modeling the datamining algorithms. We take the data mining modeling onlyconcerns with algorithms introduced by Fayyad shown in Fig.3.
4.0 INFORMATIONRETRIEVAL (IR) i
In IE, we retrieve only structured data. And IR provides alltypes of information access such as analysis, organization,storage, searching and retrieval of information. As acsordingto Slaton's classic textbook: "Information retrieval is!a fieldconcerned with the structured, analysis, organization, storage,searching, and retrieval of information" [5]. I
Information retrieval is defined by Carlo Meghini et al.I
"Information Retrieval as the task of identifying documents in
a collection on the basis of properties described to thedocuments by the user requesting the retrieval" [4].
Combination of IR and database will be valuable for thedevelopment of probabilistic models for integrityunstructured, semi-structured and structured data, for thedesign of effective distributed, heterogeneous informationsystems [5]. Jimmy Lin describes the information retrievalcycle as given in figure4 .
The Information Retrieval Cycle
Source~1>,!I'.nn
Vocabularylearning, Relevance
Source reseleclion
Figure 4: Information Retrieval Cycle
In figure4, source selection is first step in IR cycle.Appropriate resource is selected for all valuable data. Thenquery will be formed according to user needs in queryformulation. Query will be processed for searching desiredresults, rank list will be prepared according to search resultsand appropriate documents are selected. If selection is notaccording to user needs, selection criteria goes back to sourceselection and query formulation. Examination process checksthe documents for validation, if its validation is proved thendocuments will be the result otherwise output of examinationgoes back to source reselection and query formulation.
Various challenges in IR:New challenges to IR community and motivated researchersto look for intelligent Information Retrieval (IR) systems thatsearch and/ or filter information automatically based on somehigher level of understanding are required. What type ofsemantics is to be used to improve effectiveness in IR? Whatwill be the hypothesis to improve representation of documentsand by incorporating limited semantic knowledge? [7].
5.0 INTELLIGENT INFORMATION RETRIEVAL SYSTEMS
JIANG Xinhua et al described that the retrieved documentshave the right terms but may not be in the desired context. It
I
has been argued that exploiting the user's context has thepotential to improve the performance of information retrievalsystem described in [10].
The challenge of developing intelligent informationretrieval system based on context of documents: I
• How to develop a methodology of assigning Fontextto documents using assigners and validate the'resultsthrough measurement of consistency betweenassigners?
• Web usage tools, usually implementing the process ofcustomizing the content and the structure of w~b sites
I
in order to satisfy the specific need of each and everyuser, without asking for it explicitly;
• "'Web content mining tools used for automaticclassification of document contents, including theirmultimedia objects as well as textual information.
I6.0 ONTOLOGY AND INFORMATION RETRIEVAL
JING-YAN WANG et a!. described that the characteri~tics ofan ideal information retrieval system are that searching is fleetand result is accurate described in [9]: I
Ontology is the formalized description of share confeptualmodel to the information resource, so any conceptual ~bject Ccan be defined as: C ~ {D, W, R} Where, D indicates suchdomain knowledge, W is a status set of correlative thing inapplied domain; R is a set of concept relation in domain space
~~. I
It has two effects that ontology is used in infor.mationretrieval.
1. Automatic analysis to the domain attribhte ofdocument.The information retrieval system looks for keywordsaccording to. a certain means in searched inforinationdocument; these keywords can be used in documentclassification which is dependent on ontologyknowledge. So information retrieval system can sortdocuments to some possible domains combining withprimary content of information document~ thenfiltrates off irrelevant domains, gets final re~ult ofcorrelative domains which the document is classed to.
2. Intelligent formulation and visualization to' user'ssearching demand.
It is very important that how to visuali~e user's knowledgedemand by appropriate method in preliminary s~age ofinformation retrieval.
ABOUT THE AUTHOR:
7.0 CONCLUSION
Data warehousing is a process of centralized data. management and retrieval. Data mining is the process ofanalyzing data from different perspectives and summarizing itinto useful information. It allows users to analyze data frommany different dimensions or angles, categorize it, andsummarize the relationships identified The Data MiningOntology is a formal description of the domain concepts andthe relation between the concepts, it uses a hierarchicalstructure containing the concept entities and their relation.Ontology would be understandable for both human beings
and machine and could be reused.Information retrieval provides all types of information
access. Concept-based visual indexing has high expressivepower, which can easily communicate with user but stillinvolves information loss in transforming visual materials intotext, and requires more intensive human labor. InformationExtraction is a logical step to retrieve structured data and theextracted information. The extracted information is then usedin searching, browsing, querying, and mining. The retrieveddocuments have the right terms but may not be in the desiredcontext. It has been argued that exploiting the user's contexthas the potential to improve the performance of informa honretrieval systems.Tnere are two important issues to be addressed. First one is
to design and integrate an efficient ontology and the second isto establish of linkages between ontologies of different
domains.
REFERENCES:
[1] Qi Luo, "Advancing Knowledge Discovery and Data Mining", 2008 IEEE
DOI10.1109/WKDD.=.153[2) Mao-Song Un, Hui Zhang. and Zhang-Guo Yu, "An Ontology for
Supporting Data Mining Process" Manuscript received June 30, 2CTh.[3] AnHai Doan, Raghu Ramakrishna, and Shiva Kumar, "Managing
Infom:tationExtraction", SIGMOD J.Xl6, ACM 1-59593-256-9jlli[4] CARW MEGHINL FABRIZIO SEBASTIAN!, AND UMBERTD
STRACOA, "A Model of Multimedia Infomtation Retrieval", Journal of
theACM, Vol. 48, No. 5,September2OOl, pp. 909-WO.[5] G. Salton, "Automatic Information Organization and Rebieval", Mt£raw-
Hill, New York, 196ft[6] Otallenge; In Information Retrieval and language Modeling. Report of a
Workshop held at the center for intelligent Infomtation Retrieval, University
of Massachusetts Amherst, and September 2002[7] Jimmy Un, "An introduction to infomlation retrieval and question
answering", =.[8) Tanveer J Siddiqui "Intelligent Techniques for Effective Infomtation
Retrieval", 2lXl6.[9] jING-YAN WANG, ZHEN ZHU, "FRAMEWORK OF MULTI-AGENT
INFORMATION RETRIEVAL 5)SfEM BASED ON ONTOr=Y ANDITS APPliCATION" 978-14244-20964/08/$25.oo@2008lEEE
[10] JIANG Xin-Hual, UU Yang-Min, "A New Artificial Intelligent InfommtimRetrieval Methods" 97l><J-7695-3559-3/09$25.oo@2009IEEE
[11] Jimmy' Un, "An introduction to information retrieval and question
answering",2CXJ4.[12] Chen-Yu Lee ,Von-Wun SQ(), " OLtology based Information Retrieval and
Extraction" IEEE 0-78ffi-8932-il/OS 2005 .[13] Hao Han and Takehiro Tokuda , "A Method for Integration ofWeb
Applications Based on Information Extraction" IOVE.2008. 1EEE DOl
lO.1109/IC[14] Sebastian A Kyos~ Juan D. Vei'asquezt, Eduardo S. Veraj:~, Hiro;l1i
Yasuda_ and Terumasa, " Improving Web Site Content Using a Gmcept-
based Knowledge Discove.y" lEEE/WIC/ ACM International Confereoce
on Web Intelligence 2CXXi.[15] Gilles Nachouki, "A Method for Infornlation Extraction from the Web"o-
7803-9521-2/lli/$20.oo)2006IEEE.[16] Hao Han and Takel1iro Tokuda, "A Method for Integration ofWeb
Applications Based on Infonnation Extraction".[17] Xianming Uu , Wemua U , Shixian U3 Zhenwen Tao, "A New Model to
Describe Information Semantic" International Sympc6ium on InformationScience and Engieering,2008lEEE
(18] Yi )(iao, Ming Xiao, Fan Zhang. "Intelligent Information Retrieval Model
Based on Multi-Agents" 14244-1312-5/W /$25.oo@=IEEE[19] Mohan S. Kankanhalli and Yong Rui, "Application Potential of Multimedia
infom",tion Retrieval" IEEE I Vol. 96,No. 4, April 2008 0018-9219/$25.00_2008IEEE
[20] Pan Ying.Wang Tianjiang. Jiang Xueling. " Building Intelligent InformationRetrieval System Based on Ontology" 14244-1135-1/W /$2S.oo@2007IEEE.
Poonalll Yadav obtained B.Teell in ComputerEngg.&Science frOIll Kumks/ldra UniversityKllrukslletra, India and M.Tech in InformationTec1mology frOIll Glint Govind Singh IndraprastltaUniversity in 2002 and 2007 respectively. SlJe ismrrenfly working as Assistnnt Professor in D.A. VCollege of Engg. & Tee/lI1ology, Kanina(Molrindergnrh). Presently slle is pursuing Iler PhD.researc11at NIMS University, Jaipllr. Her researc11interests include Illformation Retrieval; Web based retrieval and Semantic Webetc. Mrs. Yadav is a life time memner of indian Socieh) for Tee/mical Educationand ',er email idis:[email protected].
Dr. R. P. Singll received B.Tech. and PIID. degreesill electrical engineering frOlll K.N.l.T; Sllltmrpllr(U.P.) and Institute of Tecllnology, Banaras HinduUniversity, Varanasi (U.P.), India, in 1988 and 1993,respectively.Dr. Singll is a persall of multidimcnsional personalitywit1l23 years of expericnce in tltefield of teaching andresearch incll/ding 4 years as Director! Principal. Dr.Singlr is a member of various amdemic societies of national and internationalrepute. He has to Iris credit 40 researe/I papers and authored fOllr books.Tlrescbooks are well received by tile students & amdemicians alike. III 1994-1995 hewas tlEerecipient of "Tlte President of India" as Director! Principal aliJardforresearch contributions. Currently Ite is working as Director! Principal in Bint1aDevi Edllmt;onal Society's Group of Institutions (Integrated Campus) JBKnowledge Park, Faridabad and I,is email id is: [email protected].
