Confidence-based Link/Attribute Inference based on Friendship Circles

• Main Authors: Ke Chun Yu, 柯君郁
• Other Authors: Lee Sing Ling
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/57967134107402495139

碩士 === 國立中正大學 === 資訊工程研究所 === 102 === In this thesis, we discuss how to determine the order in recommending new friends or new interests by the confidence index generated from SVM on the social network. The main idea is analyzing ego's friendship circles and finding interests or friends, which are not currently included in ego's profile, and then score order by SVM confidence is used to decide the recommendation in every turn based on the ego network structure. The most different place is not using fixed order but score order by SVM confidence to decide the recommendation in every turn. Because fixed order may make some wrong recommendations and affect the next recommendation with a low accuracy result. In our experiment, we show that the accuracy of SVM confidence score order is better than that of fixed order in general, even when the information is not enough.