2.3 User Modeling in Information Retrieval

In IR the user model is used to provide assistance to the user in the query formulation process. The goal is to express the informational need in the best possible way. Clearly the best way is the one that provides the system with enough information to retrieve all documents that are relevant to the user. Unfortunately, the result of a search can only be as good as the description of the items that are sought. Users frequently have a hard time explicitly specifying what exactly they are looking for. It is the task of the user modeling component of the IR system to automatically complement the users' interests based on their previous search behavior.

The user model as maintained by the IR system thus augments the search specification issued by the users. The expanded search specification then hopefully reflects more closely the users' needs and interests. Fig. I.8 illustrates the central position that user modeling can assume in information retrieval systems.

Figure I.8 User Models in Information Retrieval (from [Kok91])

An IR system enhanced with user modeling techniques will normally start by getting the user's preferences. These can, e.g., be a statement of the user's interests, a self description, or an SQL-based query. This input is subsequently analyzed, using the user model, and the user model is updated accordingly. Then the formal query is constructed and processed, based on the user's preferences. Afterwards, in close interaction with the user model, the output is prepared for presentation to the user (e.g., results of the query sorted based on user's interests) and the user model is refreshed. Finally, the user can evaluate the query, and restart the whole cycle if needed.

Users must be able to communicate their wishes to the system in form of interests, while the system needs a way to return to the users its predictions of what the reflected interests in the system are. This model is now illustrated in the next section by an actual example.

---