Topic based photo set retrieval using user annotated tags

  • Authors:
  • Sangjin Lee;Jonghun Park

  • Affiliations:
  • Dept. of Industrial Engineering, Seoul National University, Seoul, Korea 151-744;Dept. of Industrial Engineering, Seoul National University, Seoul, Korea 151-744

  • Venue:
  • Multimedia Tools and Applications
  • Year:
  • 2013

Quantified Score

Hi-index 0.00

Visualization

Abstract

As a storage and retrieval unit of user generated web objects, set has been receiving increased attention recently in information retrieval research community. Set search requires relevant sets to be retrieved to meet information needs of users. It is different from individual object search in terms of content granularity. While a web object itself is not divisible and independent with each other, a set consists of separable objects that are related in some aspects. This paper proposes a new approach that can effectively measure topical relevance of sets against a user query by utilizing the tags attached to web objects. The main idea of the proposed approach is to prefer the set which covers as many query related subtopics as possible. In particular, in order to compute the topical relevance while addressing the problem of noisy tags, the notion of tag significance score is introduced based on tag co-occurrence frequency. We consider a problem domain of photo set search at flickr.com where individual photos are annotated with texts such as titles and tags. Experimental results show that our proposed method outperforms the previous approaches for photo set retrieval.