Time-Sensitive Web Image Ranking and Retrievalvia Dynamic Multi-task Regression
به نام خدا
Title: Time-Sensitive Web Image Ranking and Retrievalvia Dynamic Multi-task Regression
Authors: Gunhee Kim, Eric P Xing
Abstract: In this paper, we investigate a time-sensitive image retrievalproblem, in which given a query keyword, a query timepoint, and optionally user information, we retrieve the mostrelevant and temporally suitable images from the database.Inspired by recently emerging interests on query dynamics ininformation retrieval research, our time-sensitive image re-trieval algorithm can infer users_ implicit search intent betterand provide more engaging and diverse search results ac-cording to temporal trends of Web user photos. We modelobserved image streams as instances of multivariate pointprocesses represented by several different descriptors, anddevelop a regularized multi-task regression framework thatautomatically selects and learns stochastic parametric mod-els to solve the relations between image occurrence prob-abilities and various temporal factors that influence them.Using Flickr datasets of more than seven million images of 30topics, our experimental results show that the proposed al-gorithm is more successful in time-sensitive image retrievalthan other candidate methods, including ranking SVM, aPageRank-based image ranking, and a generative temporaltopic model.
Publish Year: 2013
حامی دانش بومی ایرانیان