A keyword based video advertising system
Poster at IEEE Automatic Speech Recognition and Understanding Workshop 2011, December 11-15, 2011, Hawaii
Contextual online advertising is widely used today. As opposed to textual information, the huge amount of video data hardly contributes to the contextual information yet. Mei and Hua therefore see the necessity for "a new advertising generation dedicated to media" and propose a second generation supporting contextual multimedia advertising. They present a system detecting ad insertion points in videos as well as providing multimodal ad matching, first to decrease the intrusiveness and latter to increase the contextual relevance of an ad in audio-visual content. For our demo we followed the suggestion of Mei and Hua and developed a new advertising system using keyword information extracted from video content. We consider descriptive or strategic keywords for adverstisements, which are used to find relevant videos and the position in a video to place the advertisement. Spoken term detection (STD) based on a large vocabulary continuous speech recognition (LVCSR) system provides the content information necessary for our keyword based approach. The system adapts to common standards and architectures for online advertising and enables online selection of contextually relevant ads.