Presentations

This study investigated the contribution of dataset reuse to the diversity of research areas using articles that reused datasets. The Framingham Heart Study (FHS) and Atherosclerosis Risk in Communities Study (ARIC), which are widely used in life sciences, were selected as samples. Disease names in the articles that reused these datasets were extracted based on Medical Subject Headings (MeSH) descriptors. The frequencies of disease names were examined from the 1950s for FHS and the 1980s for ARIC. The amount of “cardiovascular diseases” research was decreasing, while that of “pathological conditions, signs, and symptoms” was increasing from the beginning. The Herfindahl-Hirschman Index (HHI) scores, which are an index of diversity, were calculated based on disease names; these scores indicate that research areas have diversified over time. This study found the HHI and MeSH descriptors could be used to measure the contribution of dataset reuse to the diversity of research areas.

In this paper, we describe methods of classifying Japan MARC records to class number of Nippon Decimal Classification. We compare the performance of three categorization method, based on mutual information (MI), relative frequency and SVM. In each method, training data are title and table of contents and blurb on the flap in Japan MARC records. The experimental results show that the best performance is MI using title and table of contents and blurb on the flap, but other methods are not. In failure analysis, we found the performance depends on subject of class number.

コレクション : 国立国会図書館デジタルコレクション > 電子書籍・電子雑誌 > 学術機関 > 学協会

In machine learning study, we need to prepare test collections for conducting text categorization experiments. However, it has been frequently pointed out that constructing labeled data set is expensive and / or time-consuming. The purpose of this study is automatically identifying and collecting academic articles in Japanese PDF files on the Web. Then, we conducted the automatic detecting experiment using pooling method and compared the performance of various classifiers. Results confirmed applicability and usefulness of pooling method in this experimental environment.