Current thumbnail selection techniques for video ads are based on content, e.g., saliency maps, faces, logos, etc., and quality, e.g., blurriness, presence of video-encoding artifacts, etc. While such techniques perform better than random selection or default selection of the first frame, these don’t generally account for user activity or preferences. This disclosure describes techniques that leverage statistics of prior user interactions with call-to-action (CTA) buttons within video advertisements to guide the selection of thumbnails for the video ads. Thumbnails are selected from segments of a video ad where prior viewer data indicates a tendency to click the CTA button within the ad.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.