Computational Advertisement of Yelp
The auction theory is prevalent in the realm of mobile app advertisement. In this blog, I am going to expand the theories from the lecture to the real life example, using one of the largest crowed-sourced recommendation apps, Yelp. Before starting the in-depth case study, we need to clarify some definitions. In the mobile app advertisement, the owners of advertisement represent the buyers. For every user search, the buyers participate the auction to compete for the advertisement slots, which represents the items they are buying. Some items are more expensive; for example, the first banner that comes on top of the search result is more expensive than the small windows on the side. The winner of auction will pay the cost for the advertisement, and their advertisement will be displayed on the page.
When users of Yelp search for new information, the process of the auction starts. First step is to filter the candidates for the auction participants. To maximize the outcome, Yelp has an algorithm to go through its data base and select positional candidates for the auction. Many factors are considered in this process. For example, the study says that user’s interests decreases exponentially as the distance from the advertised business increases. Yelp intentionally chooses local business for the auction candidates. Also, the search category is an important factor. The challenge is that the similarity of categories does not necessarily mean the better candidate. For example, even though Karaoke and Sushi might have a complete different categories, their advertisements perform well for each others. Yelp has developped its own algorithm to maximize its revenue from advertisement.
After selecting the candidates, the auction starts. Using machine learning, Yelp calculates the expected click rate of each advertisement and its placement. Also, they decide the cost of each advertisement according to the monthly budget of clients. Finally all advertisements are sorted in the order of the product of expected click rates and the price that the company will pay for each click. The advertisement which would generate the highest revenue will win the auction and pay the cost of the runner-up advertisement. The reason why Yelp uses 2nd price audition is to encourage each companies to bid truthfully since the dominant strategy is to do so.
Even though the auction is done, there are some more factors to be considered. For example, users tend to dislike the repetition. Even if the algorithm expects the highest click rates, they do not want to use same advertisement over and over again. Also they don’t want to use up the budget of the company with popular advertisement. Since each companies pays Yelp for monthly advertisement budget, Yelp can no longer use the advertisement once they go over the budget. In order to prevent this, Yelp needs to spread the popular advertisement throughout month so that they don’t use up the budget in the beginning of the month.
As the case study shows computational advertisement involves many outside factors. There are some factors that I did not mention in this blog such as seasonal trends. Even though people tend to hate advertisement, there are so much thoughts behind this technology.