Model-driven indexing: Indexing by ignoring content

Indexing can be viewed is the process of segmenting a large information with the indices acting as descriptors for the subspaces. In the case of information retrieval systems, the indices are used to retrieve documents based on a user query, whereas in the case of knowledge navigation systems, the indices circumscribe browsable spaces of documents. In this paper, I take the position that traditional approaches to indexing are "datadriven"; an alternate "model-driven" approach to indexing has considerable promise, particularly within an enterprise setting. Traditional approaches to information retrieval assume that the most valuable indices for a document can be derived from the content of the document. While statistical techniques attempt to derive the content from statistically significant keywords, AI techniques go one step further to derive content through parsing and inferencing. In either case, the indices are completely determined by the documents in a document collection. I refer to this approach as "data-driven." When a user attempts to locate information from a document collection, the collection itself is only one part of the equation, with the user forming the other part of the equation. For the user, the effectiveness of an indexing scheme depends on how well it supports his/her goals, task requirements and level of expertise, not on abstract measures of effectiveness such as precision and recall. An alternate approach to indexing, therefore, is to start from the user’s end, deriving indices based on user’s mental models and task models. I refer to this approach as "model-driven’." In the most general case, model-driven approaches are difficult to instantiate for large document collections because of different mental and task models for different users. However, the problem takes on a different hue when we consider information retrieval and navigation within an enterprise setting. Unlike the general populace, mental models and task models exist within