Interactive social group recommendation for Flickr photos

  • Authors:

  • Zheng-Jun Zha;Qi Tian;Junjie Cai;Zengfu Wang

  • Affiliations:

  • School of Computing, National University of Singapore, Singapore;Department of Computer Science, University of Texas at San Antonio, USA;Department of Computer Science, University of Texas at San Antonio, USA;Department of Automation, University of Science and Technology of China, China

  • Venue:
  • Neurocomputing
  • Year:
  • 2013

Quantified Score

Hi-index 0.01

Visualization

Abstract

Social groups on photo sharing Websites, such as Flickr, are self-organized communities to share photos and conversations with common interest and have gained massive popularity. Currently, users have to manually assign each photo to the appropriated group. Manual assignment requires users to be familiar with existing photos in each group. It is intractable and tedious, and thus prohibits users from exploiting the relevant groups. For solution to the problem, group recommendation has attracted increasing attention recently, which aims to suggest groups to user for a particular photo. Existing works pose group recommendation as an automatic group prediction problem with a purpose of predicting the groups of each photo automatically. Despite of dramatic progress in automatic group prediction, the prediction results are still not accurate enough. In this paper, we propose an interactive group recommendation framework with Human-in-the-Loop. Given a user's photo collection, we employ the pre-built group classifiers to predict the group of each photo. These predictions are used as the initial group recommendations. We then select a small number of representative photos from the collection and ask user to assign the groups of them. Once obtaining user's feedbacks on the representative photos, we infer the groups of remaining photos through group propagation over multiple sparse graphs among the photos. We conduct experiment on 30 Flickr groups with 239,700 photos. The experimental results demonstrate that the proposed framework is able to provide accurate group recommendations with quite a small amount of user efforts.