For a user who is unfamiliar with a target domain, the first step to conduct an exploratory search task is to go over a learning phrase, which means to learn from the search results to acquire basic domain knowledge. Since lots of search results could be returned by a search engine, and usually only a small portion of all the results contain valuable knowledge to the current search task, the user usually needs to read lots of documents and could only learn limited knowledge. This makes the learning phrase a low efficiency, time consuming and easy to fail process. In order to support the learning phrase of the exploratory search process, this paper proposes the concept exploration model which describes how a user reads search results and figures out interesting concepts. The model focuses on how does a user explore related concepts during the learning phrase, and factorizes the concept exploration process as a production of the probability that concepts form a specific relation structure, and the probability that a user is attracted by a concept. In an application example, the concept exploration model is used in a query recommendation task to support exploratory search. We demonstrate how to determine the two probabilistic factors and evaluate the model with a set of metrics. The experiment results show that the application example could help users explore domain concepts more effectively.
