A Dynamic Scene Recognition Method for Event-Based Social Network

Haidong Kang, Tianhan Gao, Nan Guo

Abstract: With the development of event-based social network, the ways and tools of people’s online and offline social activities have changed dramatically. IEBSN emphasizes the impromptu of social events, which will lead to inaccurate and inefficient event recommendation. To address this problem, we propose a novel dynamic scene recognition method for IEBSN, which combines supervised learning and unsupervised learning. Aiming at the similarity and inconsistency between classes in scene recognition, we propose a convolution feature encoder, which can extract more scene visual information. In order to meet the unexpected requirement of application level, the images whose convolution module is lower than a certain threshold are clustered with k-means. Experiments show that this method can automatically identify the scene in IEBSN application efficiently, and alleviate the problems in scene recognition.