Learning Hidden Relations between People to Discover Communities in Photos

• Main Authors: Shang-Chi Chen, 陳上祺
• Other Authors: 鄭卜壬
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/66542835054307219810

碩士 === 國立臺灣大學 === 資訊工程學研究所 === 101 === The photographic technology is bringing rich nultimedia content which contains a lot of social information to us. If we can use those rich information to discover the communities that hide in the photos, we can provide a novel way to browse and organize our album and improve the socail websites’ current function. For achieving the goal, this work divids the photos to multiple socail events, because socail events are the natural and basic unit for photo collections. And this paper propose a framework that learn each event’s trustworthiness and relations which hide in those events. Then, multiple clustering algorithms can be applied and discovery the commmunities in the photo collection. In the experiments that base on real world data, we can know that user feedbacks have significant improvement on the performance of community detection. However, the clustering method and the feedback strategy should be suitable, or the proformance will be hurt. Moreover, this work also analysis the impact of tunable variables in the framework, and the result is reasonable.