Advertisement selection for online videos

The significant increase of the audience of video-sharing websites has attracted large investments on contextual advertising. In this paper we investigate alternative techniques for selecting ads to be placed in videos during their exhibitions. In order to improve the effectiveness of these ads avoiding the high cost of image processing, we explore textual metadata created by the users and stored in the page that hosts the video. Our proposal consists in assigning weights to the metadata attributes, according to their relative importance. To estimate such an importance, we apply a block importance model. Our experimental results show that the metadata attributes which perform better in the task of selecting ads are the ones that provide the best descriptions of the video content. We also observed that the application of weights according to the studied block importance model obtained gains of about 7\% over the selection method without weights. Such a gain is relevant due to the possibility of increasing the profitability of ad selection systems and avoiding the negative impact on advertiser brand and credibility of placing non-relevant ads.