权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Deepening of Recommender Systems with Explanations and their Quality Evaluation

推荐系统的深化及其质量评价

基本信息

批准号：
21H03553
负责人：
岡本一志
金额：
$ 11.32万
依托单位：
The University of Electro-Communications
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (B)
财政年份：
2021
资助国家：
日本
起止时间：
2021-04-01 至 2025-03-31
项目状态：
未结题

项目摘要

本研究は，動的・複数項目の評価データ・ユーザIDの有無が混在といった，より高次の情報や複雑な状況を対象に，推薦理由を説明する情報推薦システムを開発し，理由の説明機能の品質評価法を確立する．本年度の実績は次のとおり．（１）一定期間に発生した行動やイベントの集合であるセッションを対象とした推薦システムを，分散表現されたユーザとアイテムを近傍探索するアプローチで開発した．具体的には，Word2Vecによりアイテムの分散表現を獲得し，その分散表現の階層化・集約することでセッションとユーザの分散表現を逐次的に獲得する手法を提案し，k近傍法による推薦システムに適用した．3種類のデータセットを用いた推薦実験から，GRU4Recと同程度の推薦精度を有し，数百倍高速に学習できることを検証している．年度末には研究成果をまとめ，国際学術論文誌への論文投稿を行った．この開発はセッションデータを対象とした例示による説明機能の実現に係るものである．（２）これまでに展開してきた価値観モデリングの研究を推進し，価値観モデリングのアプローチで複数項目の評価データを対象とする推薦システムの開発に取り組んだ．具体的には，価値観をユーザの個別の評価項目（属性）へのこだわり度と仮定し，ラフ集合を用いてユーザが付与した属性評価値と総合評価値の関係をIf-Thenルールで抽出し，そのルールに基づき推薦するシステムを提案した．Livedoorグルメデータセットを用いた推薦実験より，従来のユーザベース協調フィルタリングと比べ，推薦精度が向上することを確認している．この開発は複数項目の評価データを対象としたルールによる説明機能の実現に係るものである．

In this study, the complex items involved in this study are not related to each other. In this study, the complex items involved in this study show that there is no error in the ID. The reason for this is that the reason is that the reason is clear. The reason is that it is possible to make sure that the law is in place. (1) during a certain period of time, health and health activities are collected. (1) during a certain period of time, health activities are collected in a certain period of time. They are similar to each other, and are scattered to show that they are close to exploring the market. Specific information. The Word2Vec distribution system shows that the performance is satisfactory, and the dispersion table shows that the data collection is effective, and that the data collection is analyzed in detail. The method proposal is proposed step by step. In the near future, it is recommended that the accuracy is improved in the same way. 3. The accuracy is different in the same way. At the end of the year, there was a review of the research results and the submission of articles in the journal of national science, and there was a bank review on the contribution of articles and articles in the end of the year. This is an example of an example of how your computer can realize the performance of your computer. (2) the development of research in the field of high-speed communication. This is not true because the number of complex items is different. The number of complex items is similar to that of the database. For a specific example, you can select the number of items (properties) that are unique to each item (attribute), and the collection of data sets can be used to extract the data from the If-Then. In order to improve the accuracy of the system, we need to make sure that we are sure that we are in a good position to make sure that we are in good agreement with each other in order to make sure that we are in good agreement with each other. In this way, we can make sure that the error is not correct. Turn on the complex item to verify that the message is correct.