S-Space College of Engineering/Engineering Practice School (공과대학/대학원) Dept. of Computer Science and Engineering (컴퓨터공학부) Theses (Ph.D. / Sc.D._컴퓨터공학부)
Emotion Based Item Recommendation Techniques in Social Cataloging Services
소셜 카탈로깅 서비스에서의 감정 기반 아이템 추천 기법
- 공과대학 전기·컴퓨터공학부
- Issue Date
- 서울대학교 대학원
- Social Cataloging Service; Connected Users; Isolated Users; Recommendation; Tag; Emotion; Tensor; High-Order Singular Value Decomposition; Probabilistic Ranking
- 학위논문 (박사)-- 서울대학교 대학원 공과대학 전기·컴퓨터공학부, 2017. 8. 김형주.
- Social cataloging services allow users to catalog items, express subjective opinions, and communicate with other users. Users in social cataloging services can refer to others activities and opinions and obtain complementary information about items through the relationships with others. However, unlike a general social networking service where user behaviors are based on the connections between users, users in social cataloging services can participate and contribute to services and can obtain the information about items without links. In contrast to a general social networking service in which actions are performed based on connections between users, You can
participate and contribute. In this doctoral dissertation, we classify users into two groups as connected users and isolated users and analyze usersbehaviors. Considering the characteristics of users who mainly focus on contents rather than relationships, we propose a tag emotion-based item recommendation scheme. Tags are the additional information about the item, and at the same time, it is a subjective estimation of users for items, which contains the users feelings and opinions on the item. Therefore, if we consider the emotions contained in tags, it is possible to obtain the recommendation result reflecting the users preferences or interest. In order to reflect the emotions of each tag, the ternary relationships between users, items, and tags are modeled by the three-order tensor, and new items are recommended based on the latent semantic information derived by a high order singular value decomposition technique. However, the data sparsity problem occurs because the number of items in which a user is tagged is smaller than the amount of all items. In addition, since the recommendation is based on the latent semantic information among users, items, and tags, the previous tagging histories of users and items are not considered. Therefore, in this dissertation, we use item-based collaborative filtering technique to generate additional data to build an extended data set. We also propose an improved recommendation method considering the user and item profiles. The proposed method is evaluated based on the actual data of social cataloging service. As a result, we show that the proposed method improves the recommendation performances compared to the collaborative filtering and other tensor-based recommendation methods.