A method is disclosed for media content selection based on historical presence duration. The method includes running a MapReduce model over ad visibility logs and building a key for each ad impression opportunity. For each key, a model is built that describes the visibility distribution over time. The MapReduce calculates the percentage of impressions that pass the visibility time threshold and stores it in a data store. At ad serving time the advertising service determines the minimum view-through rate time threshold for each candidate advertisement. The method then involves calculating the likelihood that the user will view the advertisement long enough. This factor is then included into the view-through rate prediction, which is factored into the auction bid, and the media content with the best likelihood of being viewed on an application or web page wins the auction.

Creative Commons License

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