application/pdf

インターネット上で提供される講義資料や講義内容のアーカイブ化が進むと,目的の資料や授業を探すための検索が必要なとなる.しかしながら,このような複数のデータからなる複合データは,その内容を単純な語句により説明することは極めて困難であり,既存のリレーショナルデータベース等の技術をそのまま利用していては,目的のデータを発見することは容易ではない.本稿では,このような複雑な特徴を有するデータからなるデータベースの効果的な検索のために,我々が開発を進めている,媒介変数を利用することにより個人の興味,関心,嗜好の分析をおこない,これをデータベースの検索結果に反映させることで,利用者が目的とするデータの発見を支援する手法について述べ、提案手法に基づくデータベースアプリケーションの構築と、これを利用した評価実験の結果を紹介する.

In this paper, wedescribe a newretrieval method we are developing. in order to find target dataefficiently from databases, which are consisting of some indistinguishable data by specific words because of their complex characteristics. It could reflects individual preferences of users on retrieval results using mediation variables, and facilitate users to find target or preferable data. Based on our methods, we design and implement an on-line recipe retrieval system as an example of database applications, perform an experiment with the implemented system, and show the effectiveness of our methods with the experimental result. Now, we are archiving lecture data for Internet School of Tohoku University (ISTU) in order to provided them via the network. Some multimedia databases are constructed on the web, and are providing their data (e.g. movies, music and others). The archives grow lager, some effective retrieval methods become to be necessary to find target data. However, the characteristics of the contents that are consisting of complex data such as the lecture data, movies data, music, and other multimedia data are hard to be expressed by some specific words. It makes difficult to distinguish different data and find target data from archives when we are using some existing relational database or full-text search methods as it was. In order to find target data efficiently from such complex databases, we are proposing a new retrieval method. This method uses the other data as mediation variables, which have some relation to target data. It defines mapping function from target data to mediation variables in advance based on the relation betweenthem. Itestim ates individual interests, concerns or preferences of differentusers through the mediation variables, and uses estimated preferences as evaluation functions for retrieving. Thus, it reflects them on retrieval result, and facilitate users to find target or preferable data, although there is not some specific words to distinguish target data. Based on the proposing method, we designed and implemented an on-line recipe retrieval system as an example of web based database application. It estimates users' individual preferences for dishes through wine types as mediation variables based on our method, and recommends preferable dishes to an userwhen he

she queries by materials of dishes. And, we performed experiment using implemented system. We prepared two systems; one is the system without our method, and the other hand is implemented with our methods, and compared effects between them. As a result, we found that subjects could find target data efficiently by the system with our method compared with the system without our method, and confirmed the effectiveness that our method could reflect individual preferences on retrieval results.

紀要類

論文

980286 bytes