演題番号 : P3-q18
臼井 支朗 / Shiro Usui:1 Nilton L Kamiji:1 谷口 達樹 / Tatsuki Taniguchi:2 上田 修功 / Naonori Ueda:3
1:理研・脳センター・ニューロインフォマティクス / Lab for Neuroinformatics, RIKEN, BSI 2:（株）IVIS / IVIS Inc., Tokyo 3:NTT-CS, Kyoto
Searching relevant abstracts submitted to scientific conferences in this field is becoming more and more challenging as the society grows bigger and bigger. For example, Japan Neuroscience Society meeting in 2009 had over 1,600 abstracts, and this number is increasing year by year. For this year conference, it is a tri-society meeting, therefore even more presentation is expected. In this work we will present a tool for searching relevant abstracts based on document similarity between three abstracts, that is, a triplet similarity. Here, document similarity is calculated on-line to allow users search for abstract related to multiple selection simultaneously, providing a larger degree of freedom compared to other tools. It does not require any thesaurus, and it also has features for suggesting possible keywords candidates to help user refine their search. Moreover, the Neuroinformatics Congress 2010 will be held at the same venue just prior to Neuro 2010, therefore, cross-searching between the two meetings will also be made available.We have evaluated the effectiveness of the triplet similarity compared to the ordinary cosine similarity measure, and have demonstrated that the most important abstracts are listed on the top of the list. Surprisingly, half of the abstracts selected by the users were those that did not contain the keyword the user has originally entered, which we call "hidden treasure". In other words, by only doing a keyword search, the user would never find those abstracts unless the correct keywords were entered. Users can also create their own list of abstracts by adding important abstracts to the printing cart, which is automatically sorted according to presentation time together with keywords utilized for the search to help users organize their search result. The printing cart can be accessed for review and print from any computer on the web. Try and find your hidden treasures!