In an academic paper search, particularly a search to confirm the originality of a us-er’s research and to create survey articles, it is important that the search returns com-prehensive results related to the user’s information need. [1] proposes a method for efficiently selecting relevant research papers from a vast abstract set, which is based on a topic model and search formula created by the user. In this paper, we construct a system that visually expresses categories included in the reduced set of papers from [1], using a clustering based on the user’s information need and selecting clusters having relevant papers. To generate the clusters based on the user’s information need, it is important to know which structures (such as “background” and “method”) in the abstract are relevant to the information need. This is based on the knowledge that if a user searches the papers related to the automatic construction of a thesaurus, he/she will judge whether a paper is relevant from sentences in the abstract describing the research purpose and method. We therefore propose a method using only the sentence content that matches the information need in the clustering.