Recently, development of search engines helps to simplify human's information seeking behavior, because we can easily access useful information from the Web. However, search engine users tend to select only Web pages in a search result returned by the search engines considering their rankings. Consequently, the users just browse highly ranked Web pages, and cannot be conducive to find useful information because of turning their eyes away from the identification of their information sources, which is the most important task in the information seeking behavior. In order to solve this problem, people should use the mechanism which can show procedures to get useful information, and should emulate the capabilities of searchers in public libraries using their affairs. In this paper, we propose a method for extracting user's information seeking behavior to develop the mechanism. In order to extract the information seeking behavior, we use the collaborative reference database developed by the National Diet Library (NDL) of Japan as searchers' behaviors, and analyze its dialogues called "reference data."