The goal of the algorithms for the HP web content identiciation challenge is to retrieve or label all the informative multimedia content in web pages. The performance of the algorithms would be measured by comparing the automatically computed results with the manually labeled ground truth. Ground truth is generated by manually labeling all the informative multimedia content (i.e. images/video/flash objects) in a pre-selected set of the web pages (in various languages). The algorithm is expected to retrieve nearly all the informative multimedia content in the web pages.
The precision is defined as the number of informative images classified/labeled correctly by the algorithm divided by the total number of images labeled as informative by the algorithm. In other words, precision is the number of true positives divided by the sum of true positives and false positives. Recall is defined as the number of informative images classified/labeled correctly by the algorithm divided by the total number of informative images (which should have been labeled as informative). Recall is the number of true positives divided by the sum of true positives and false negatives. The final comparisons between the algorithms will be made by computing the F-measures using the precision and recall.
Here are examples of the types of web pages that we plan to use to evaluate the submissions:
http://www.mapquest.com/maps?1c=Palo+Alto&1s=CA&2c=San+Francisco+&2s=CA (Driving directions)
(Note, images of similar products or other product recommendations are considered informative)
http://news.sina.com/oth/phoenixtv/502-104-103-108/2009-05-27/01323899156.html (entertainment articles)
http://blog.naver.com/honeykja/40045216645 (blog, recipe)