Dynamic Ontologies on the Web
Jeff Heflin and James Hendler
Department of Computer Science University of Maryland College Park, MD 20742 {heflin, hendler}@cs.umd.edu

Abstract
We discuss the problems associated with managing ontologies in distributed environments such as the Web. The Web poses unique problems for the use of ontologies because of the rapid evolution and autonomy of web sites. We present SHOE, a web-based knowledge representation language that supports multiple versions of ontologies. We describe SHOE in the terms of a logic that separates data from ontologies and allows ontologies to provide different perspectives on the data. We then discuss the features of SHOE that address ontology versioning, the effects of ontology revision on SHOE web pages, and methods for implementing ontology integration using SHOE’s extension and version mechanisms.

1. Introduction
The World Wide Web is a repository of information that is structured for presentation to human readers and is thus mostly inaccessible to machines. This situation will be somewhat alleviated by the Extensible Markup Language (XML), which allows content to be separated from presentation. However, although XML Document Type Declarations (DTDs) can specify the grammar of markup languages, there are no facilities for formalizing the meaning of these languages. To create a web language with semantics, one must extend XML with features of knowledge representation (KR) languages. However, the Web presents new challenges for KR; simply creating an XML syntax for traditional KR languages is insufficient. The Web is a distributed system and there are many providers of information. As such, the reliability of information is questionable, and it is inevitable that inconsistencies will arise. The Web is also in a constant state of change. Although standard vocabularies will be necessary for interoperability, these vocabularies must be able to evolve as the Web does. Additionally, the sheer size of the Web will test the scalability of KR systems and algorithms. As a consequence of the Web’s size and dynamic nature, it must be treated as an open-world, since it would be Copyright © 2000, American Association for Artificial Intelligence (www.aaai.org). All rights reserved.

unrealistic for any agent to assume that it knows all true facts about the Web. The Simple HTML Ontology Extensions (SHOE) is an ontology-based knowledge representation language that is embedded in web pages (Luke et al. 1997; Heflin, Hendler, and Luke 1999). Over the course of four years, we have investigated the use of ontologies to support the structuring and querying of data on the Web. We begin this paper with an overview of the SHOE language, and then provide a logical semantics for it. We then discuss the problem of ontology revision, which is necessary in a dynamic environment such as the Web, and describe how SHOE’s versioning mechanism copes with this. Next, we discuss the tendency of distributed ontologies to diverge, and provide methods for reintegrating them using SHOE’s extension and version mechanisms. Finally, we discuss related work and present our conclusions.

2. SHOE
The underlying philosophy of SHOE is that intelligent internet agents will be able to better perform their tasks if the most useful information is provided in a structured manner. To this end, SHOE extends HTML with a set of knowledge oriented tags that, unlike HTML tags, provide structure for knowledge acquisition as opposed to information presentation. SHOE associates meaning with this content by making each web page commit to one or more ontologies. These ontologies permit the discovery of implicit knowledge through the use of taxonomies and inference rules, allowing information providers to encode only the necessary information on their web pages, and to use the level of detail that is appropriate to the context. Interoperability is promoted through the sharing and reuse of ontologies. To achieve compatibility with existing web standards, SHOE’s syntax is defined as an application of SGML, a language that defines tag-based languages and was the influence for HTML’s syntax. A slight variant of the syntax exists for compatibility with XML, and can be used by web sites that have begun to migrate to XML. The nature of the SHOE language makes it possible to develop numerous tools and architectures for processing it. In order to evaluate the language, we have built a suite

... ... ... ... ... Figure 1. An Example Ontology of tools, including a tool for adding SHOE markup to web pages, a web crawler that gathers SHOE from web pages and stores it in a knowledge base, and a number of query tools. We have applied these tools to various domains, including computer science departments and food safety. A discussion of these tools and applications can be found in Heflin, Hendler and Luke (1999). Demos of the tools are available at http://www.cs.umd.edu/projects/plus/SHOE. Ontology reuse in SHOE is accomplished by extending general ontologies to create more specific ontologies. Specifically, the tag indicates the id and version number of an ontology that is extended. An optional URL field allows systems to locate the ontology if needed and a PREFIX field is used to establish a short local identifier for the ontology. When an ontology refers to an element from an extended ontology, this prefix and a period is appended before the element’s name. In this way, references are guaranteed to be unambiguous, even when two ontologies use the same term to mean different things. By chaining the prefixes, one can specify a path through the extended ontologies to an element in a general ontology. An ontology can define categories, relations, and other components. Categories are introduced with a tag and may specify one or more subsuming categories. Note that it is not possible to specify subsuming categories for a category defined in another ontology. Relations, which are essentially n-ary predicates, are defined with a tag and must specify types for each argument. Sometimes an ontology may need to use a term from another ontology, but a different label may be more useful within its context. The tag allows the ontology to specify a local name for a concept from any extended ontology. This local name must be unique within the scope of the ontology in which the rename appears. Renaming allows domain specific ontologies to use the vocabulary that is appropriate for the domain, while maintaining interoperability with other domains. SHOE uses inference rules, indicated by the tag, to supply additional axioms. A SHOE inference rule consists of a body of one or more subclauses describing claims that entities might make and

2.1 Language Features
In this section we describe the features of SHOE that are necessary for an understanding of this paper. The reader may be surprised by the simplicity...

-14 -31 -32 -42 -441 -62 -66 -71 -74 -97 -98 -99 /projects/plus/shoe. /tr/1999/pr-rdf-schema-19990303.html. /tr/rec-rdf-syntax19990222.html. 0 1 16 18 1990 1994 1995 1997 1998 1999 2 2.1 2.2 2000 20742 25 3 3.1 3.2 36 4 4078 436 5 53 56 59 6 aaai abl abstract accept access accommod accomplish achiev acknowledg acquisit across activ actual ad add addison addison-wesley addit address adjust advanc advantag af_ont.fighterpilot affect age agenc agent ai al algebra algorithm alia align allevi allow also altern although ambigu american amsterdam analysi ancestor and/or anoth answer anteced antiqu appear append appli applic approach appropri approv arbitrari architectur area argument ari aris arithmet ariti armi art artifici ascertain aspect assembl assign associ assum assumpt atom attempt author automat autonom autonomi avail axiom axioms1 backward backward-compat backward-compatible-with base becam becom begin begun belong benefit berlin better bind bodi bottom brickley broker build built busont1.employee busont2.staffmember byproduct c ca call cannot car care categori central certain chain challeng chang chosen ckml claim clarif classifi claus clear collabor colleg come commit common communiti comparison compat complet complex compon composit comput concept conceptu concern conclus concurr confer conflict conjunct consequ consequence2 consid consist consortium constant constraint construct contain content context contract contribut conveni coordin cope copi copyright core correct correspond cost could cours crawler creat criticont1.rating criticont2.rating cs cs-tr cs.umd.edu current cyc d daal01 danger darpa data datalog deal decad decid decker declar deepli defens defin definit delimit demo depart depend dept describ descript design detail determin develop differ difficult direct disadvantag discov discoveri discuss disjoint distinct distinguish distribut diverg document domain done draft drop ds dtds due dynam earlier easi ed effect effort either element elimin els embed enabl encod end entiti environ equival erdmann error essenti establish et evalu even everi evolut evolv exact exampl execut exist expect expens explicit express extend extens extent extrem f faa_ont.jetpilot facil fact fail far far-reach farquhar featur feet fensel fewer field figur fike file final first first-ord fit flow focus follow foo food form formal formed former formula found four fourteenth framework fulfil function fundament furthermor futur g gather general giaretta give given good govern grammar greater ground grow growth guarante guarino guha guid hafner handl head heflin hendler herbrand heterogen horn howev html human id ideal ident identifi if-and-only-if ignor ill ill-prepar illustr impact implement implicit import imposs improv inaccess includ inclus inconsist incorpor indic individu inevit infer influenc inform inherit initi instanc instead insuffici integr intellig intend intent intern internet interoper interpret intersect introduc introduct investig invis involv io irrelev isol issu item ith j jame jeff k0135 kent key keyword know knowledg kr l label laboratori languag larg lassila last later lattic least legaci lenat less let level like littl local locat logic longer loosen loss luke m machin machine-read machineri made magazin main maintain mainten make malici manag mani manner manual map mar markup maryland mass match matter may md mean meant mechan member membership menlo mere merg mess meter method might migrat miniatur model modif modifi modular monoton monterey most move much multipl must n n-ari name namespac nation natur naval necessari necessarili need negat network never new next non non-log none note notion noy number numer ny o o1 o2 object obviat obvious occasion offici old om omit oml one ontobrok ontolingua ontolog ontology-bas open open-world openworld opinion oppos option order orient origin os other otherwis overlap overview own owner p pace page paper paraphras park parti particular path perform period permit perspect phase philosophi pi place point popular portion pose possibl postgradu practic pre pre-process precis predic preexist prefer prefix prepar preprocess present press prevent previous primarili principl problem proceed process professor program project promot proof properti propos provid public publish purport purpos queri question quick r r.e r.v rager rapid rate rather rdf reach read readabl reader reason recommend reduc refer reflect reintegr relat reli reliabl reload remov renam replac report repositori repres represent request requir research reserv resolv resourc respect rest restrict result reus revers revis revised-bi rice richer right rule safe safeti said say scalabl scale scenario schema scheme school scienc scope second section see seem semant semi semi-automat sens sentenc separ serv server set sgml share sheer shoe short show shown side similar simpl simpli simplic sinc singl single-valu site situat size slight solv somebodi sometim somewhat sourc specif specifi spector springer springer-verlag standard state statement static step stop store structur studer subclaus subset substitut subsum succinct suffici suggest suit supercategori suppli support surpris swick symbol syntax synthes system t1 tag tag-bas take task taxonomi technic tend tendenc term terminolog test theori therefor thing third thought three thus time tn togeth tool top toward tr tradeoff tradit translat treat trivial true tupl twelfth two type typic u.s umiac unambigu unari unavoid unchang under understand undiscov unintend uniqu univers unless unlik unrealist unsuit untam unwieldi upgrad upon url use user usual v valid valu var variabl variant various verlag versa version vice view virtu vocabulari w3c want way weak web web-bas weigh well well-form well-formed wellform wesley whatev whether wide wiederhold wilder wish within without work workshop world world-wid would ws www www.aaai.org www.cs.umd.edu www.cs.umd.edu/projects/plus/shoe. www.w3.org www.w3.org/tr/1999/pr-rdf-schema-19990303.html. www.w3.org/tr/rec-rdf-syntax19990222.html. x x1 xi xml xn year yet york æ