Review Analysis WWW2012 Weinan Zhang 29 Feb. 2012
General Info Acceptance Rate: 12% (108/885) Monetization Track – Gui-Rong Xue – About 60 submissions – 4~5 accepted papers
Two papers Paper 301: Joint Optimization of Bid and Budget Allocation in Sponsored Search – Internet Advertising Team, MSRA Paper 324: A Semantic Approach to Recommending Text Advertisements for Images – ApexLab
Paper 301 Joint Optimization of Bid and Budget Allocation in Sponsored Search – Sponsored Search Advertiser-Oriented Service
Solution Probabilistic Model for Ad Ranking Joint Optimization on Bid Price and Campaign Budget Experiment on Simulator
Review Comments RatingConfidence Borderline (0)Medium (2) Borderline (0)High (3) Weak accept (1)High (3)
Pros Interesting and important problem Real auction data Good written
Cons Budget constraint The optimization problem and solution are straightforward The experiment is only a simulation
Sum up of paper 301 Three times – SIGIR, WSDM, WWW – More than 10 footnotes now Unsolved points – Straightforward model – Simulation – Value per click estimation Submit to KDD
Paper 324 A Semantic Approach to Recommending Text Advertisements for Images – Cross-media Mining – Thesis of bachelor – First submission
Visual Contextual Advertising
Our Solution JeepCar Auto Vehicle Plane Truck
Review Comments RatingConfidence Weak Reject (-1) High (3) Weak Reject (-1)Expert (4) Weak Accept (1)High (3)
Pros Semantic match outperforms syntactic matching Interesting – “The idea is very interesting and I would love to see this as a full paper ” but…
Cons Image and ads may not match any concept – Even Wikipedia is not sufficient Part of ads collection is retrieved by WordNet words Matching between knowledge bases is trivial in this paper Should provide more detailed results – Accuracy of each node of ImageNet
Sum up of paper 324 Adding knowledge bases – Wikipedia – More LOD here – Folksonomy Not just knowledge bases – Image: Image annotation, ViCAD – Text Ads: Bid Keywords Deeper experiment results Plan to WSDM
Lessons Learned More detailed experimental results – Accuracy of locating nodes in Imagenet for input images – Effectiveness of different matching functions More non-experiment efforts – Discussion – Writing
Thank you