Graphical Table Of Contents - Drexel University

2m ago
955.31 KB
9 Pages
Last View : 2m ago
Last Download : 28d ago
Upload by : Duke Fulford

GraphicalTable of ContentsXia LinSchool of Library and InformationUniversityof KentuckyLexington,KY 40506xlin@ukcc.uky.eduABSTRACTThis paper proposes a graphical table of contents (GTOC)that is functionallyanalogous to the table of contents. Theproposed GTOC can be generated automaticallyfrom thetext of documents.It visualizesdocument contents andrelationshipsto allow easy access of underlyingdocuments.It alsoprovides various interactivetools to let the userexplore the documents.Issues of how to generate suchGTOC include how documents are indexed and organized,how the organizeddocumentsare visualized,and whatinteractivemeansare neededto providenecessaryfunctionalityof GTOC. These issues are discussed in thispaper with a GTOC prototypebased on Kohonen’s selforganizing featore map algorithm.1.INTRODUCTIONWhat kinds of formats that the table of contents needs to bein the digital environment?This paper proposes a graphicaltable of contents (GTOC) that is functionallyanalogous tothe table of contents in the printed environment,Like thetable of contents, GTOC will provide (1) an overview to thecontents covered in a book or a collectionof articles, (2) aset of semanticclusters (or sections)that group relateddocuments together, and (3) a quick access to documents bytheir contents.Furthermore,because of its associativeindexing and its visual features, GTOC will encourage usersto browse or scan informationby their visual perception,whichis the most effectivemethodfor receivingandassimilatinginformationfor humans (Arnheim,1971).The proposed graphical table of contents will be generatedautomaticallyfrom the text of documents,either from abook, one or several issues of a journal,or a set ofdocuments in a database. Issues of how to generate suchGTOC include how documents are indexed and organized,how the organizeddocumentsare visualized,and whatinteractivemeansare neededto providenecessaryfunctionalityof GTOC.In this paper, these issues areelaborated using examples of GTOC generated by a trate that semantic structures of documents can beabstracted and visualized to assist informationaccess in thedigital environment,and graphical displays of documentswill likely support those functions of the table of contents.In the next section, literatureof contents will be reviewedrelated to research on the tablefkst.Permission to make dlgitellhard copies of all or part of Wla material forperaoml or claaamam use ia gmnted without fee provided that the copiesare not made or dktributed for profit or cornnwvialadvantage, the. copyright notice, the title of the publication and ita date appear, and notice isgiven that copyright is by perrniasion of the ACM, Inc. To copy otherwise,to repubtish, to peat on servers or to mdiatribute to fiats, requires apeeificpm-mission andlor fee.DL’96,Bethesda MD USA@1996 ACM ()-89791-83&4/9(j/03. . 3.502. FEATURESOF THETABLEOF CONTENTSAlmost every book or every issue of a journal has a tableof eontents (TOC).Surprising y, there has been much lessresearch on TOC than on other components of documentssuch as titles or title pages, abstracts, and back-of-thebook indexing.Whileit is difficultto contradicttheimportance of TOC, it seems to be a common sense thatuse of TOC is rather simple, and constructionof TOC isstraightforward.This is perhaps true in the printedenvironment,but this will certainly not be the case in theupcomingdigitalinformationenvironment.In thissection, features of TOC will be reviewed, and how thesefeatures should be implementedin the digital environmentwill be discussed.The organizationof TOC in the printed environmentmayseem to be trivial:for every book, or every assembledcollectionof documents,titles and authors of individualdocumentsor chaptersare listedalongwiththeirpagination,which,all together,becomethe table ofcontents.Nevertheless,practices of TOC are varied frombooks to books, and journalsto journals.Juhasz, et al.(1973)analyzedmore than 120 primaryjournalstoinvestigate features of TOC. Major variations were found interms of title listing and pagination system, the sequentialarrangementof author, title, and pagination,and leadersbetween paginationand other elements.These differencesindicate that the visual appearance is a major feature of TOC.In fact, if such a comparisonis done between books orscholarly journalsand magazines, the visual differenceiseven more obvious.For example, one can easily tell adifference between TOC of scholarly journals and TOC oftrade magazines: while the formal uses exclusivelytext in arather formal way, the latter has a variety of text, picturesand other display elements.Anotherfeature of TOC ia to providean overviewtocontents of books or journals.TOC groups articles orchapters into sections with section headings.The groupingalso separates differenttypes of articles such as “researchpapera” and “book reviews. ”People can get to knowcontents of a book by scanning through its TOC. They canquickly judge whether or not any chapters or articles of thebook are of interest to them. Prabha et al. (1988) surveyed331 library users on the use of non-fictionbooks.Theyfound that TOC was the feature most commonlyused todetermine which books to borrow.Because of this overviewfeature, TOC is also a good tool for current awarenessservices. Researchers often browse through TOCS of someselected periodicals to keep track of recent .service.E-mail“TablesorofotherOn the Internet,TABLE-OF-

CONTENTS,Inc. (Gopher:// providesaccess to weeklyupdated TOCS of 165 top magazines,which,as they claim, “keeps you at the absolute cuttingedge of what’s happening.”Clearly, the most importantfeature of TOC is to provide adirect access to individualdocuments or book chapters.Inthe printedenvironment,this functionhas been usedeffectivelyby users. They can quicklyidentifyindividualtitles from TOC and follow the page numbers to documents.In the digital environment,however, TOC has not been wellincorporatedinto the searching environment.Research ononline catalog (Markey,1983) indicated that TOC was themost importantinformationmissingin currentonlinecatalogs and to be able to search TOC was the most desirablefunctionneeded to add to online catalogs.Thus, mostresearch on TOC is centered on using TOC for contentenriched access (Cochrane,1985, Van Orden, 1990).Theexperimentconducted by Dillon and Wenzel (1990) clearlyshows that addingTOC willimproveoverallretrievaleffectivenessof online catalogs.Current library online catalogs have successfullymodeledthe card catalog for informationaccess at the book level.However,the digital library wearebuildingnow will needaccess not only at the book level but also at the article andIn the digitalsectionlevels,even at the idea level.environment,there willbe no longer a clearlydefinedstorage format called “page.”There is even less certainwhat constitutes a “book.”A(paper)book is a collection ofrelated items, a result of a query search from a digital libraryis also a collectionof related items.Whileitems in thebook are brought together by the author, items in the resultof a query are broughttogether accordingto the reader’sneeds (the query statement).If, in the digital library, theuser can group a set of needed items to create a “dynamicbook,”and can access these items using many featurescurrentlyavailablein books, the user willlikelyfind iteasier to access the needed informationin such a “dynamicbook. ”To explorean analogversionof the printedtable ofcontents,we need to considerdifferencesbetweentheprintedand digitalenvironments,and the features of thetable of contents reviewed early.It is proposed that, in thedigitalenvironment,“ the table of contents needs to be generated automaticallyto reflect the dynamicfeature of “digitalbooks” andonlinecollections. thetable of contents needs to provide an overviewtocontentsof the documentsit covers; the overviewshould reflect how documents or key concepts in thedocuments are semanticallyrelated to each other. the table of contentsneeds to “self-organize”thedocumentsinto clusters,groups,or sections,with appropriateheadings automaticallyassignedtable of contents needs to be visuallyattractive tosupport browsingand visual sense-making.the table of contents needs to be simple and easy to use,it needs to providea directaccess to underlyingdocuments.theThe GTOC prototype described later is based on these ideas,and the idea of visualizationfor informationaccess. Thenext section reviews related literature on informationvisualization.3. VISUALIZATIONACCESSFORINFORMATIONIn the printedenvironment,people are used to seeinginformationon tangible media (such as papers and books)and stored in visible locations (such as libraries).They cando a recognitionbetter than do a recall when searchingobjects.In the digital environment,it becomes difficultto“see” informationand to “recognize”informationby visualperception.Peoplecan only rely on textual queries theygenerate to retrieve informationfrom databases.However,as the resultof rapiddevelopmentof graphicalandvisualizationtechniques,this situationis likely to change.The computer will soon create or reintroduce visual cues thatare lost wheninformationis convertedfromprintedresources to electronicresources.It willrearrangeandreassociateinformationitems to reveal new associationsthat might not be seen otherwise(Veith,1988).It willcreate visual interfacesthat allow visual and perceptualinformationseeking in the digital environment.To reach this level of informationseeking, a critical issue ishow informationshouldbe representedand visuallydisplayed.Researchers have addressed this issue for years.Doyle (1962) suggested“semanticroad maps” based onword associations that could provide “a view of the entirelibrary at a distance” and help the searcher to “narrow hisfocus by recognition.”Miller (1968) emphasized that whatwaa needed was “a spatialorganizationfor the storedinformationthat is more compatiblewith the structure ofthe informationitself.”Sammon (1969) mapped a set ofdocumentsby a nonlinearmappingalgorithmto revealdocument associations.Fairchild et al. (1988) proposed andtesteda three-dimensionalnetworkknowledgebase,SemNet, where the user can view both local and globalstructures of the associative network.Fowler et al. (1991)applied a mathematicalmodel, “the PathfinderNetworks, ”to unify a visual space for queries, documentsand terms.Arents and Bogaerts (1993) explored a structure called “Cubeof contents” to visualize structures of hyperindexing.Themost comprehensiveinformationvisualizationprojectisthe InformationVisualizerdevelopedin XeroxPARC(Robertaon,et al. 1993).The ionformats,suchasDataMap,InfoGrid,ConeTree,and Perspectivewall,tovisualizeinformationfrom differentperspectivesand atdifferent levels of abstraction.In a series of research, Lin (Lin, et al., 1991; Lin, 1992)proposed a map display that would show both contents andstructures of a document space. The map display representsa “survey” of all the documents in the document space. Itdefines a spatial analog for the documents,and revealscontentsand semanticrelationshipsof documentsbyvarious visual cues such as distances, links, clusters, areas,and neighborhoods.The underlyingengine for the mapdisplay is a neural networks’ learning algorithm,Kohonen’sfeature map (Kohonen,1989).Because of the algorithm’slearningand self-organizingfeature, the map display canmaintaindocument semantic relationshipsas measured byword occurrence and co-occurrence,and it can show majorcontent areas that “win” over other areaa through recursivecompetitions.The map display was evaluated by comparingits structuresand functions to some human-generatedmap displays (Lin,The map displaywas akw tested in anet al., 1993).46

experimentthatinvolvedsixty-eightconducted some simple retrieval tasks withsubjectswhothe map display(Lin, 1995).These studies contributedto an understandingof how informationshould be organizedand displayedgraphically.It was during these research activities that theidea of GTOC sparkled: if a map display was generated basedon one or several tables of contents,the displaymightfunctionas a table of contents in the digital environment.Thus, an experimentalGTOC prototype was implementedtostudy functionsand features of the table of contents in thedigitalenvironment.The next sectiondescribestheprototype.4. THE GTOC PROTOTYPEThree proceduresare designedto generatethe GTOCprototypean indexing procedure, amappingprocedure, anda displayprocedure.The indexingprocedureconvertsdocuments to numerical vectors. The procedureis currentlybased on the VectorSpace Model(Salton,1989).Theprocess includes:(1) extract a list of words from thedocuments,(2) delete stopwordsfrnm the list, (3) use aword-stem procedure to reduce the list to stem form, and (4)create indexingvectors based on the stemmed list.Theindexingvectors can be either in a binary form based onword occurrence,or in a weightform based on wordfrequencies and/or inverse document frequencies.Once theindexing vectors are generated, the mapping procedure willstart to organize the indexing vectors and associate them toa visual space in the end. The procedure currently is basedon Kohonen’sself-organizingfeaturemap.It createsneighborhoodson a two-dimensionalspace where similardocumentsare mapped to nearby locations.Finally,adisplayprocedurewillplot the mappingresults on aninterfaceand link variousinteractivefunctionsto themappingresults.Figure 1 shows a GTOC for SIGIR proceedings 1986-1993,for a total of 292 documents.Statically,the display showsa general overviewof the contents.All the areas, theirlocations,and their labels, are automaticallydrawn by thedisplay procedure based on the mapping results. The largerthe areas, the more often the (labeling)terms are discussedThe closer the two areas, the morein the proceedings.likelythe terms in the two areas are co-occurredindocumentsof the proceedings.Interactively,the displaycan reveal even more information.The slider on the bottomof the map controlsthe number of terms shown on thedisplay.When the slider is moved to the left, only thoseterms with high “activationlevels” are displayed, when theslider is moved towardthe right, keywordswith lower“activationlevels” are added to the display.Because theactivationlevel of a term indicatesfrequenciesof termoccurrenceand co-occurrence,those with high “activationlevels” can be regarded as “major” terms in the collection,which are shown on the interface first. Figure 1 shows theSIGIR GTOC in two levels: first, major areas are identifiedby termssuch as “systems,”“fulltext .As the slider is moved to the right, moreterms are added to the display. These terms are shown on thedisplayat the locationsdeterminedby their associations.They improvedescriptionsof the areas and clarifysometerms shown early. For example, for the “model” area, therelated terms are “vector space,” “extended boo lean,” andwhichrepresenta good sampleof IR“probabilistic,”For the “queries” area,models discussed in the literature.47words added to it are ion,”and “process.” These terms again show strongsemantic relationships.When the slider moves to furtherright,more terms willbe added to the display,andrelationshipsof areas or terms will become increasinglyclear.GTOC users can decide at what level of details the displayshould show. They will typicallystart to browse the displaywith a few terms on it. If a term or an area is perceived to berelated to their informationneeds, they will naturallyfocuson that area of the display when adding more terms to thedisplay.If the display becomes confusing,they can alwaysmove back the slider to reduce the number of terms on thedisplay.When one or several terms are identifiedto beassociated with the needed information,they can click onthat location, a pop-up windowwill show the top 10 titlesassociatedto this location(Figure2).These titles arearranged by their associativeweights,and they are likelysemanticallyrelated to the terms clicked.When users find atitle they are looking for, they can click on the title to seethe full display of the article, just like they would turn tothat page when a title is found in the table of contents.Ifthey do not find the title, they can adjust the clickinglocation to open another (title) window.After they movearound the surface for a while, the structures of the displaywill become visuallyfamiliarto them, and it will becomemuch easier to click on the right locationfor documentsthey are looking for.A major difference of the term display and the title display isthat while each term is mapped to a unique location on thedisplay,titlesmay be mappedto multiplelocationsdynamicallybased on their associationsto the terms.Therefore, users do not need to click on an exact location fora title, clickinganywhere withina neighborhoodof a keyconcept in a title will “fire up” the title.If the title hasmultiplekey concepts or key terms, it will likely appear inThis feature tolerates“fuzziness”ofmultiplelocations.visual perceptionso that users only need to decide anapproximatelocation for a title they look for and the titlewill be shown up there very often.5. ISSUESOF GTOC CONSTRUCTIONThe GTOC prototypewas constructedas an experimentaltool to study new forms of the table of contentsin thedigital environment.During the prototypedesign process,many theoretical and practical issues occurred.These issuesare essential to the concept of GTOC,and much moreresearch is needed to study various problems related to theseissues. In this section, our recent research effects on theseissues are described, and some primary results are presentedand discussed.5.1 DocumentindexingrepresentationanddocumentIt is important to realize that any visual display can only beas good as the input used to generate the display.Thus, howthe underlyingdocuments are represented and indexed willhave a significantimpact on the final organizationandviews of GTOC. The Vector Space Model used in the aboveexample is widelyacceptable in the informationretrievalHowever,there are stillmanydifferentcommunity.approaches to apply the model to practical problems.Thereare also differentconsiderationswhen applyingthe model

Tableet hodlinguisticContentsThis is an associative mapdisplay for SIGIRproceedings 1986-1994.queriesWords on the map (andtheir relationships)represent contents of llanguagelctionshypertextseurusf Ullor the location. )iniwfaceTo add more words on themap, move the slider to theright. To reduce thenumber of words, move theslider to the left.Click on my word orlocation to see titlesassociated with the wordIructienti?xtClick on any title to viewfull record of thesystemsdesignccessofstructueindexing document,1expansGi—networkskrmIIlogicstud decum(,c i.lwTT”Trl-rl r nowledoereprwentetionsIll I4“interfaceintelligentproblemsu3e rdev; lv;mttext.wmprswiwnexperts !.I l ,:nsdesignIIi ndexiwarehFIGURE 1. Dynamic views of GTOC. GTOC presentscontentsin different levels ofdetails. As the slider moves to the right, more terms are added to the display based on theirassociative relationships.48

Gr8phicalTablekethod.4——.i rnprovedlinau!s icstructuet-queriesIi mediadII--l I-placationsq“ Scatter{Gather:ACluster- BasedAooroachto BrowsinoLaOnthe Allocationof Documentsin Multiprocessor Info;rna “ConstantInterection-Time Scatter/GatherBrowsingof WHierarchic DocumentClustering UeingWard’s MethodUser-OrientedDocumentClustering: A Frameworkfor LetIntegrating Query,Thesaurus,ati Documents through a C #j!.trm ion““.seurwiExperience with Large Oocument Collectionsi lj;Structured Answers for a Large Structured Documant CoilBExtmrimentswith CIueruAcquisition andUsein Document Qusersystemsdesign“mlt lof Contentfi ndexiIIaccess 8activities3adaptive 3advanced 2algorithms6rinrilysis12answering 3applications8approaches automating 3bibliographic3bit 2bitmaps 2boolean 4browsing 6case 2classificationcluster 13cognitive 32Figure2. Pop up windowsof GTOC.While the map display shows a general overviewcontents, ‘the ‘POPUDwindow shows titles related to the cl cked location iin this examole.word “ciuster” ‘is l[cked).for retrieving and matching purposes and for visualizingandbrowsing purposes.For example, using title words only forindexingis usuallyconsideredas a limiteddocumentrepresentationfor informationretrieval.Using words fromtitles,abstracts,and fulltextto indexdocumentswillimprovethe power of representationsignificantly(Salton,1989).On the other hand, dimensionsused to index thecollectionare also significantlyincreasedwhentheindexingchanges from title words only to words fromabstracts or fulltext.For visualizationpurpose, words usedfor indexing are also candidate words for labels on the visualdisplay.As the size of display space is limited, the numberof labels that can be put on the display is very limited.Thisleads to the question of how much visual improvementitwould make when the indexing changes from a much highdimensionaldocumentrepresentationbased on fulltextindexingto a low-dimensionalrepresentationbased ontitles.We may expect that, while using every word in thefulltextimprovesthe representation,words appear in thetitles may still be the most useful words to be used as labelsin a visual display of documents.Thus, it is necessary to compare visual displays generatedby different indexing procedures in order to determine whatindexingproceduresto applyfor GTOCconstruction.Currently,we are comparingthree differenttypes ofindexes:49ofthe(1) TT-indexing:words that appear in at least two titlesare collecte as the based et for indexing,after thestopword-removing and stemmingprocedures,avector is created for each document using the binaryrepresentation,that is, a “ 1” is used if the wordcorrespondingto the particulardimensionisappeared in the title, and a “O” is used otherwise.The based set of words used for(2) TF-indexing:indexingis collectedthe same way in the TTindexing,but the indexingprocedureis based onwordsin titles,keywords,andabstractsofdocuments.A vector is created using the weightrepresentationof euses the samelowdimensionsas the title indexing,but the indexingvectors reflect how the indexing words are distributedin titles, keywords and abstracts.(3) FF-indexing:Every word from titles, keywords andabstracts is collected.Those that occur in at leastthree documents are used as the based indexing set. Avector is created for each document using the nt frequencies (idf).This is a standard fulltext indexing procedure.

Number ofdocumentsNumber ofTermsTraining8) Among the top 20% (as measured by the activationlevel) of terms in the third display, about 80% ofthem appear in the titles.In other words, if thedisplaycan only show 20% of fulltextindexingterms due to the limiteddisplay space, title wordscan account for most of them.TimeTT-indexing143126TF-indexing143126351 seconds352 secondsFF-indexing1435681908 secondsTablel.Descriptive data onapplyingthethree pesofindexingforKohonen’amapping.The training time refers to 2500 iterations ofKohonen’smappmg on 14 by 14 output nodes,doneon a convexmachine.Figure3 givesthe threevisualdisplaysof SIGIRproceedings1990-1993,based on the three differenttypesof indexing,and Table 1 gives the descriptivedata ofapplying the three types of indexing procedures to generatethe GTOCdisplay.An immediatequestionis how tocomparethese displays?Even thoughthey are visualrepresentationsfor the same documentset, they arecertainlydifferent“views.”On one hand, because differentmeasurements are applied, we shouldn'texpect the displayswould look the same. On the other hand, because the threeindexingmethods all generate representationsof the samedocument set, we should expect to see similar patterns ofterm relationshipsor document relationships.Two approachesare currentlytaken to comparethesedisplays, one is a visual inspectionto examine patterns ofterms and documents, the other is an experimentalstudy totest how the displays help users find documents. The visualinspectionwas done throughinteractionwithall thedisplays.It is difficultto describesimilaritiesanddifferencesof these displaysin snapshots such as thoseshown in Figure 1 and 2. However, once we interact withthe displays, using the sliders and the popup windows,thefollowingtrends seem to be cleac1) Words always used together in this collection,suchas “relevance feedback” and “minimumperfect hash(functions),”are always mapped together (note thatall the indexing procedures are based on individualwords).2) Words often used together, such as “interface design”and “full text,” are always mapped closely nearby(not togethersometimes,because words such asare also used in some other“design”and “text”contents).3) Semanticallyrelated words, such as “explorations”and “visualization,”or “terms”and “phases,”areoftenmappednearbyor withinthesameneighborhood,even though they may not be usedtogether.4) Local neighborhoodrelationshipsseem to reflectsemantic relationshipsreasonablywell (judged bythe document contents) in all the three displays, butthe overall layouts of the displays are all different.5) When same words are clicked on the three displays,titles shown in the popup windows sre similar, butnot the same (typically,the first few titles are thesame and thev contain the clicked words).0)“’ The inverse d cument frequencies “punish” the highfrequencywords, makingthe visual displays less“visuallyintuitive”(in terms of large areas forand small areas for lesshigh-frequencywords,frequency words).the numberof indexingterms increases7) Whensignificantly(as in the third display),all the areasbecome small and similarin size, which makes itdifficultto spot terms visually.These observationsare useful guidelinesfor the design ofGTOC prototype.Observationsnumber1, 2, 3 and 5establish some validityof the displays,or create certain“trusts” that users can rely upon what they see for what theyare looking for. Number 4 cautions that general overviewsof a documentcollectionmay be representedin manydifferentways graphically,each perhapswithcertain“distortions. ”Number6 to 8 questionsif the inversedocumentfrequencyor fulltextindexingis usefulforvisualizationof documentcollections,in the particularsituation we defined.Based on these observations,we hypothesizedthat TFindexing would be the best format for the GTOC prototype.The one shown in Figure 1 is based on this format.To testthis hypothesis,an experimentwas designed to comparethree versions of GTOC (lT-based,TF-based, and FF-based)and the originaltableof contentscopiedfromtheproceedings.Currently we are collectingand analyzing datafor this experiment.5.2 DocumentvisualizationThe other majormappinganddocumentfactor that affects the views and functionsof GTOC is how documents are mapped to the visual display.Document mapping is to connect “views” of the display toits underlyingdocuments.Typical tools used for documentmapping are those used for exploratorydata analysis, 8),andartificialneural networks approaches (Mao & Jain, 1995).Whilethereare manysimilaritiesbetweenmappingnumericaldata and mappingdocuments,their differencesdeserve a special attention because of unique features of theFirst, the document space is often highdocument space.dimensional,but vectors used to represent the documents aregenerallysparse.Second,the documentspace is also“feature-less”in the senses that the dimensionitselfisdefined artificiallyduring the indexingprocess.Finally,clusteringor organizingthe numericalrepresentationofdocumentsitself may not be the final goal of documentRather, the goal willbe to extract semanticmapping.relationshipsof documents and to provide enough data andto ailow people to Thus, a good mapping procedure will need to(1) preserve the inherent structure of documents as well aspos ible while projectingthe high-dimensionaldocumentspace to a two-dimensionaldisplay, (2) abstract features anddetermine how to ,qroup documents on the visual display, (3)group similardo&mentsinto clustersor neighborhoodstructures, (4) visuallyidentifyclusters, neighborhoodsandindividualdocumentsand theirrelationships,and (5)providedata for implementationof interactivetools fordynamic views of the document space.Duringthe design of GTOC prototype,many differentThese algorithmsincludealgorithmshave been ping(Sammon,1969), PrincipalComponentAnalysis(PCA)(Friedman& Tukey, 1974), and Kohonen’s self-organizing50‘

associationtrmIi braruelectronicIexi ngcom pressianexPl0ratl“Onshypertext(a)co repleteIndexed by title words only, using sqstemsi ntcrfacerel evan Cefeedbac kstructuringal gc.rithmsrrrocessinferencequeriesIi nteracttondocumentBva n“ce?6MCk1I1III essonsr--l%lmet hodnew9(b) Indexed by title words, using theiroccurrences in titles, keywords andabstracts to compute within documentfrequencies.-1’3! ----1‘yStemsIprobabilt’ ticm-- uJrl’tr”ct”ringDI i l.Ill.filesgnaturetextautomaticI SemanticinvestigationmariaiI moppl nqexampleslogicfreenetwork i ofere nceprojects(c) Indexed by every word from titles,keywords, and abstracts, using bothwithin and inverse documentfrequencies.associati0n3measurecodl ngl-i-vectorfeedbackrelevanceautomaticspace11t hcseur uscl assi fi cation0 ri e ntati 0 narchitectareIIi nguistict heoryextendvisualizationlesson

table of contents reviewed early. It is proposed that, in the digital environment, " the table of contents needs to be generated automatically to reflect the dynamic feature of "digital books" and online collections. the table of contents needs to provide an overview to contents of the documents it covers; the overview

Related Documents:

ACADEMIC ADVISOR, ACE/NACT BSN PROGRAM Margaret Breslin, BSN, RN (267) 359-5794 or Andrea Perry, MS (267) 359-5795 or ACADEMIC ADVISORS, BSN CO-OP PROGRAM Shannon Edwards (267) 359-5659 or Jayme Lloyd (267) 359-5779 or

CS 338: Graphical User Interfaces. Dario Salvucci, Drexel University. 1 Lecture 8: Design Guidelines CS 338: Graphical User Interfaces. Dario Salvucci, Drexel University. 2 The GUI development process We've now seen some tools that enable us to build our interface. So let's get back to the GUI process.

Generating Software Requirements Specification (IEEE Std. 830 1998) document with Use Cases James L. Goldman, George Abraham*, and Il Yeol Song College of Information Science and Technology (iSchool at Drexel) Drexel University, Philadelphia, PA 19104 {jimg, sga72, song} * File Size: 285KB

{me52, lb353, david} ‡School of Biomedical Engineering Science and Health Systems Drexel University ABSTRACT Cell sorting is a fundamental phenomenon in morphogen-esis, which is the process that leads to shape formation in livi

Drexel University Steven Morris . that Pearson correlation coefficients could lead to counter-intuitive results in author co-citation analysis. 2.2 Network Evolution The latest advances in statistical mechanics of complex networks have at

May 01, 1987 · of the Year award from Drexel University’s Col lege of Business and Administration. The award will be presented at the college’s Honors Day ceremony on Thursday, May 14, 2:00 P.M., in Drexel’s Main Auditorium, 32nd and Chesnut Streets. He will also be a gue

St. Katharine Drexel Church (Mechanicsburg, Pennsylvania) “The first church in the world named for Saint Katharine Drexel” 3,000 - 4,999 Cherubim We gratefully acknowledge our sponsors who have generously partnered with us, and kindly ask you to show your support to them! Richard & Mary

Drexel University Policy Year: 2021 - 2022 Policy Number: 812834 . (877) 409-7361 . This is a brief description of the Student Health Plan. The plan is available for Drexel University students and their eligible dependents. The plan is insured by Aetna Life Insurance Company (Aetna).

Nuclear Engineering (p. 52) Software Engineering (p. 56) About the College Drexel University's College of Engineering has emphasized its strengths in engineering, science and technology to train students to become the leaders of the future. In little over a century, Drexel University has

National Autism Indicators Report: Vocational Rehabilitation 3 Life Course Outcomes Research Program A.J. DREXEL AUTISM INSTITUTE, DREXEL UNIVERSITY.

The graphical desktop presents a Graphical User Interface (GUI) for the user to interact with the system and run appl i cto ns. If you hav eus dh x -b r login, you will have to start the graphical desktop manually by entering the command startx followed by the ENTER key. Fig. Starting the Graphical Desktop Note: The graphical desktop that we .

Creating a table of contents The Insert Index/Table window (Figure 1) has five tabs. All of them can be used when creating a table of contents: Use the Index/Table tab to set the attributes of the table of contents. Use the Entries and Styles tabs to format the entries in the table of contents. Use the Background tab to add color or a graphic to the background of the table of

user and the application. This kind of environments are known as a Graphical User Interfaces (GUI). Graphical interfaces are present in various types of devices and platforms, such as web form or a smartphone application. Most, if not all, graphical user interface based applications use an event management based architecture. Applications

Graphical User Interface Programming –GTK SEEM3460/ESTR3504 1. Backgound GTK supports the develop of graphical user interface (GUI) in Linux. GTK was adopted as the default graphical toolkit of GNOME and XFCE, two of the most popular Linux desktop environments. GT

It provides multi-factor authentication (graphical, text,POI-order, POI-number) in a friendly intuitive system [12]. Various methods of graphical Password Authentication Techniques As said earlier graphical password schemes can be grouped into three general categories: recognition, recall, and cued recall [7, 8].

UNIT 5 – Analytical Applications of Differentiation 1. State the Mean Value Theorem and show a graphical example. 2. State Rolle’s Theorem and show a graphical example. 3. State the Extreme Value Theorem and show a graphical example. 4. Define an absolute maximum (minimum) of a function and give a graphical example. 5. Define a re

Probabilistic Graphical Models Raquel Urtasun and Tamir Hazan TTI Chicago May 23, 2011 Raquel Urtasun and Tamir Hazan (TTI-C) Graphical Models May 23, 2011 1 / 30. . Graphical Models May 23, 2011 16 / 30. Bias-Variance trade o If the hypothesis space is very limited, it mig

Probabilistic graphical models combine a graphical representation with a complex distri-bution over a high-dimensional space (Koller & Friedman, 2009).There are two ad-vantages of using this model to study word order universals. First the graphical structure can rev

Graphical Models with R. Soren Hojsgaard, David Edwards, Steffen Lauritzen Probabilistic Graphical Models: Principles and Techniques. D. Koller and N. Friedman. Applications of Graphical Models . But do often have a “pre-

Massachusetts Curriculum Framework for English Language Arts and Literacy 3 Grade 5 Language Standards [L]. 71 Resources for Implementing the Pre-K–5 Standards. 74 Range, Quality, and Complexity of Student Reading Pre-K–5 . 79 Qualitative Analysis of Literary Texts for Pre-K–5: A Continuum of Complexity. 80 Qualitative Analysis of Informational Texts for Pre-K–5: A .