Multi-perspective User2Vec: Exploiting re-pin activity for user representation learning in content curation social network

Content curation social networks (CCSN) develop rapidly. Pinterest and Huaban are two typical CCSNs. Recently, there is active research on CCSNs. As a kind of content based social network, CCSNs involve not only the explicit social relations from user "following", but also content-based social relations from re-pin paths and so on. In this paper, we propose a novel user representation learning algorithm, Multi-perspective User2Vec Representation (MUVR). It combines the two types of social relations to get the rich user sequences. Then the representation learning is implemented by using the skip-gram algorithm. Experimental results on Huaban.com demonstrate that the proposed algorithm can represent network well. It presents more competitive results in the followee recommendation, re-pinner recommendation and multi-label classification.