The effects of heterogeneous information combination on large scale social image search.

• Main Authors: CHENG, Zhiyong, Ren, Jing, SHEN, Jialie, MIAO, Haiyan
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2011
• Subjects: Databases and Information Systems; Numerical Analysis and Scientific Computing
• Online Access: https://ink.library.smu.edu.sg/sis_research/1446
• Institution: Singapore Management University

This paper documents a comprehensive empirical study of the effects of heterogeneous information combination on large scale social image search. Our goal is to investigate how various kinds of information source can contribute the improvement of the retrieval effectiveness. In particular, a linear combination has been applied to merge search results from search module based on textual and visual features. Also, we propose different weighting schemes to integrate different kinds of query evidences in a nonlinear way. A series of experiments have been conducted using two large scale social image collections. Empirical results suggest that the system based on textual features yields much more effective and reliable results comparing to one using visual information. Further, the combination of two information sources can consistently enhance the final accuracy.