Capture of videos from websites is a basic work for searching video and analyzing video content. Discovery of advertising video URLs effectively and accurately is very important for video capture. It helps to improve the precision of video capture, optimize network utilization and reduce storage space. Currently, video content-based methods have a good ability to discover advertisement but in a limit speed, which do not meet the requirements of real projects. Since increasing achievements of URL based technologies have been made on classification subject of web pages, we utilize this technology to discover advertising video URLs. The method first produces a collection of URL segments. Next by applying N-gram feature selection, we get totally 2500 features. Afterwards, via combining the statistical information and selected word vector, the final features are generated. We use Naive Bayes, C4.5 Tree and SVM to train models. Ultimately, the experiment shows SVM is the most suitable model for discovery of advertising URLs discovery with 94% precision.