A framework is examined, in which the users interactively access documents, like scientific papers, with a physical structure appearing in the layout and a logical structure based on their contents. It supports effective and flexible use of the documents by allowing the users to retrieve relevant logical units through specification of their contents and/or roles in the document, and to browse those units and their contexts by strolling across both logical and physical structures. The whole framework and a method of document analysis that reconstructs the logical structure of a document and constructs its representation are mainly discussed in this paper.

This paper introduces a context search engine designed for answering trend-related queries. Aiming at narrowing the gap between user's information need and functions provided by an existing search engine, we are developing advanced search engine that focuses on the task of answering trend-related queries. As the task of answering trend-related queries is supposed to be common in various domains, we expect it could be used for various purposes. After explaining the structure and function of the proposed search engine, its potential application and the possibility of improvement are discussed.

The important thing in Exploratory Search is that a retrieving person clarifies the goal of search. For that purpose, first it is required to find the keyword which related to Search-word. Then, a retrieving person finds the related keyword that he is interested in. However, since the information acquired by search is huge, it is difficult to find the keyword which fulfills conditions from the information. Then, I thought that such a problem was solvable by developing the tool which extracts only required information from search results and displays the clustered result. In order to make a clustering result intelligible visually, a selforganization map is used, and information is arranged and displayed on a two-dimensional map. Moreover, in order to be able to reflect a user's idea in a clustering result, it enables it to change freely the parameter of the feature vector used by SOM. Finally, evaluating the usefulness of this tool by experiment.

Automatic synonym acquisition is an important problem in the field of document retrieval and data mining using natural language data. In this paper, we conducted two experiments to identify the properties of word vectors acquired by the Skip-gram model, related to the synonym identification. In the first experiment, we confirmed that the cosine similarity of a synonym pair is significantly higher than that of a non-synonym pair. In the second experiment, we show that only a limited number of components of word vectors are needed for discriminating synonym pairs from non-synonym pairs.