This article originated as member question about topic maps. The requestor wanted to know:

What's a topic map?
A simplistic definition of a topic map is an electronic thesaurus in which you get to define the kinds of relationships between the words. Imagine for a moment that a farmer and a lawyer in France decide to share an office and hire a librarian to manage their joint book collection. The farmer has an article about to grow avocados ("avocat" in French), and the lawyer has a directory of attorneys ("les avocats" in French). The librarian creates two cards in the "A" section of the subject card catalog — one for "avocat (fruit)" and another for "avocat (profession)." When the librarian decides to automate the card catalog, she could use a topic map data structure to enable the computer to tell whether a document is about avocados or lawyers.

This example illustrates three points about topic maps:

1. They are used to help computers sort out complex and/or ambgiuous semantic relationships. Humans communicating in their native language within a single social or professional culture are pretty good at it already.

2. Topic maps are useful when you're looking for information in a heterogenous electronic collection (i.e. one that deals with more than one business function, profession, industry, or language). The number of heterogenous collections has exploded with the advent of the Internet, the rise of multi-cultural governing bodies (e.g. the European Union), and the mandate for closer cooperation among government agencies since the 9/11 disaster.

3. Topic maps are behind-the-scenes data structures intended to extend the functions of the classic back-of-the-book index and thesaurus in an electronic environment. However, instead of using these familiar terms and concepts to describe them, the IT industry has borrowed a vocabulary from philosophy and linguistics — an obstacle to interdisciplinary collaboration and communicating with business managers.

What do topic maps look like?
Topic maps are designed to be "read" by a computer, not a human being. In this sense, they are different from a classic thesaurus like Roget's, which is useful as a writer's aid.

They can take many forms, depending on the software used to produce them and the application they are intended to support. See the IT views below. The user typically does not see the topic map, only its effect (e.g. a Web site navigation scheme or data entry screen). See the user view below.

Topic map: IT developer's view
The two screen shots below show what a topic map might look like to a system designer.

click image to enlarge	Topic map (graphic view). This is a graphical representation of a topic map for works of opera produced by a program called Omnigator by Ontopia.
click image to enlarge	Topic map (coded text view). This is a representation of the opera topic map in a text format called XTM — a special XML format developed for topic maps. Both of these examples were taken from the Omnigator Index of Topic Maps.

Topic map: user's view
The screen shot below illustrates what a user might see in an application supported by a topic map.

click image to enlarge

Topic map application. The IRS "Tax map" is an internal application that uses a topic map to organize information for IRS telephone tax law assistors. Compare this screen with search results on the public IRS Web site. For more on this application, see David Brown's presentation.

Another example
To illustrate the possibilities for expanding a classic A - Z index, we recently added some topic map capabilities to our Web-based Lab. If you look for articles about the Compaq computer company in the Montague Institute's A - Z index, you'll find a cross reference to Hewlett-Packard. This "see also" reference tells you that the two companies are related, but not how. However, if you look for Compaq in the "Organizations" section of the A - Z index, you'll see a cross reference that says "see also articles about Hewlett-Packard (acquired Compaq)."

click image to enlarge	Classic index entry with cross reference. If you look for articles about Compaq in the A - Z index, you'll see a cross reference (see also) to Hewlett-Packard. You know there's a relationship, but not what kind.
click image to enlarge	Index entry with topic map cross reference. If you look for articles about Compaq in the list of organizations, you'll see a cross reference (see also) to Hewlett-Packard telling you that H-P acquired Compaq.
click image to enlarge	Index entry with topic map reciprocal cross reference. If you look for articles about Hewlett-Packard in the list of organizations, you'll see a cross reference (see also) to Compaq telling you that Compaq was acquired by H-P.

Note that there is a two-way relationship between Compaq and H-P, just as there is between the terms in the classic thesaurus relationship. In other words, if "lion" is a narrower term for "cat," then "cat" must be a broader term for "lion." In the case of Compaq and H-P, the relationship is acquired and acquired by. In a topic map, the designer can invent an unlimited number of relationship types. In the example below from the Montague Institute Lab, the entity "name" can have four kinds of relationships — current name/former name, acquire/acquired by, employs/employed by, and makes/made by.

click image to enlarge

Organization name record in the Montague Institute Knowledge Base. New kinds of relationships for names (i.e. organizations, people, products) make it possible to see employees that work for Microsoft, companies that Microsoft has acquired, and products that Microsoft makes. The relationships are reciprocal, i.e. the product record for "Outlook Express" lists "Microsoft" in the "Made by" relationship.

Other topic map applications
We found the following examples of topic map applications:

• Health care (Canada) see "Applying Topic Maps to the Classification of Health Interventions"
• Medicare claims processing (U.S.) see "Can Topic Maps describe context for enterprise-wide applications?"
• Other applications at the U. S. Navy and Lexis-Nexis are mentioned in "Searching smarter, not harder."

For more applications of topic maps, ontologies, and other semantic integration technologies, see TopQuadrant's gallery of "capability cases."

How to create a topic map
Creating a topic map involves 4 steps:

1. Needs assessment. What steps are involved in the business process or function? What content is needed to complete the steps? Who are the potential users? How can a topic map make the process more efficient?

2. Topic map design. What are the key data elements? What are their attributes? What are the key relationships between them? In what context do they occur?

3. Data entry or extraction. The topic map skeleton must be populated with data, either by hand or by computer extraction from electronic sources.

4. Topic map implementation. The topic map and its data must be made available to the application. Often this means exporting the data to a relational database or an XML file.

Topic map creation and implementation are interdisciplinary activities that require the expertise of human Knowledge Base Editors as well as IT staff. They are also subject to the same political and budgeting pitfalls that are involved in taxonomy development or application integration. For comments on specific implementation issues, see this webinar Q&A from TopQuadrant, a leading topic map consulting firm.

Topic map software
It's possible to create simple topic maps by simply adding more relationships to a taxonomy created with a relational database. But for complex projects or those that need to be merged with external topic maps, specialized software is useful. Commercial software examples include:

• Ontopia based in Oslo, Norway. Offers "a complete set of tools for building, maintaining, and deploying topic map-based applications."
• Mondeca based in Paris, France. Applications include federated search, a multilingual portal, organization scheme for a document management system, collaborative knowledge base for industrial products.
• Empolis based in Gütersloh, Germany. Empolis is a subsidiary of arvato AG, an international media service company and part of Bertelsmann AG. Specializes in workflow services for the media and publishing industries.

An open source alternative is the Protege Ontology Editor developed by the Knowledge Systems Library at Stanford. The program, which can be downloaded free of charge, includes a "plug-in" (optional software module) for editing topic maps. For a much longer list of software, see Topic Map Tools.

Comments from members and others
Comments, links, and referrals in response to the question include the following:

1. "You might try contacting Vivian Bliss at Microsoft (vbliss@microsoft.com), Mike Uschold at Boeing (michael.f.uschold@boeing.com) or Sam Oh in Korea (samoh@yurim.skku.ac.kr). All have worked with topic maps and ontologies and could probably give you some good information or pointers to others with case studies. Also, do a quick search on the SIGIA-L archives to find the names of many others working in this area."

2. "You seem to be saying that "The TAO of Topic Maps" is a fair summary of what Topic Maps are. Actually, it's a fair summary of what one flavor of Topic Maps are -- the flavor that Ontopia products implement. The Topic Maps paradigm is quite general and it makes no presumption of a built-in ontology that, for example, necessarily includes "occurrences" (the "O" of "TAO"). The notion that all topic maps necessarily have occurrences is not justified by a close reading of the ISO standard, in which occurrences are entirely optional, and, in fact, are really a matter of syntactic sugar. The Topic Maps paradigm has really always been about semantic integration. This facilitation is provided, at the most fundamental level, by means of disclosing certain ontological commitments. These disclosure requirements are outlined in Part 5 of the ISO Topic Maps standard -- a portion of the standard that is still in draft form -- called the Topic Maps Reference Model."

3. Contributed links:

• Toward Coherence in Knowledge Through All Things Ontological (conference)
• Semantic Web Best Practices and Deployment Working Group
• Topic map news
• Topic map vendors
• Knowledge Technologies conference 2002

Conclusion
Topic maps evolved from classic linguistic tools like indexes and thesauri. Using the topic map structure, Knowledge Base Editors can design their own kinds of semantic relationships and implement them in a computer-readable format that is accessible to other applications. However, like single-source publishing and other technologies that initially seem to be elegant and efficient, topic map software has a steep learning curve. Few information managers know how to design topic maps and even fewer business managers understand what they're good for.

Topic maps are used primarily to enhance information retrieval and reduce the cost of integrating data from multiple applications in business or e-commerce applications. We are now seeing real world applications in publishing, government, and business along with specialty consulting firms and software vendors. However, standards are just beginning to emerge, and topic maps are only one of two ways to represent complex semantic relationships (the other is the Resource Description Framework or RDF).

We would do well to ask ourselves how important topic maps would be if we did a better job on the front end of information management — higher quality content, more careful content selection and editing, more effective employee training, more capable desktop knowledge management tools, and better enterprise management. In other words, how often do we really need to put the farmer and the lawyer in the same office? If we have better front end information management, topic maps that do fill a real business need will be easier to create, less costly to maintain, and make a bigger contribution to the bottom line.