This presentation will focus on lessons learned from almost six years of building and maintaining a major topic map as a contractor to a national government agency. The topic map and the methodology used to produced it have evolved continuously. Along the way, there have been some interesting surprises.

The most challenging part of this application is that we are not permitted to change any existing practices or procedures in the agency’s information creation and management procedures. We have used a bottom-up approach that enables us to extract, integrate, and revise information emanating from source documents over which we have no control or influence.

Topic mapping is used as a design principle to aggregate all information available about given subjects. It is not (yet) used to interchange information between applications.

Some of the strict distinctions between topic names, occurrences, associations have been blurred. Some documents (for example, forms) are considered both sources for occurrences and topics. Another example is the distinction between associated topics and occurrences. For the sake of presentation, related topics are displayed as if they were occurrences.

We started naturally to exploit index markers as input for the topic map. But we had to change our approach because of side effects due to integrating many indexes, namely that the use of indexes produced a topic map that the customer did not like, with too many topics, too many distinctions without differences, and suboptimal usability. We switched to using the section headers as sources for topic names, and we added a revision mechanism to transform them into strings suitable
for direct use of topic names.

The approach we are using is a constant monitoring of what we get by extracting information from documents that customer is constantly updating and then tweaking the results. We have implemented mechanisms to revise and edit the integrated information, and this mechanism feeds the automatic process of building the topic map. The tweaks are always evolving, and their effects must be monitored. The data that drive the tweaking mechanism are maintained independently of the documents.

A new requirement has arisen: keeping track of the way the topic map is built, to be able to account for the existence of every individual topic and association between topic. The questions we have to account for include: where does this topic come from? Has this topic been deleted and what was the reason for it? In order to be able to answer this kind of questions, we have devised a new model and new tools that provide us with this information.