Auction Algorithm for Distributing Ads to Customers
https://web.stanford.edu/~saberi/adwords.pdf
The study above was done by a group at Stanford University on a new potential algorithm to maximize the matches that happen at online auctions for ads. The problem is defined as the Adword problem, where given a specific number of bidders, each with a specific budget for a set of query words, the algorithm is supposed to maximize the revenue for the ad company. Because of the instantaneous nature of the user loading the webpage and the ads displaying, the algorithm also needed to take as little time as possible.
Companies auction to be able to advertise to certain customers by betting on keywords, and can also set daily limits on how much they want to spend advertising. The auctions are all done automatically and are similar to a sealed bid as the companies have already decided on and sent in the value that particular words are worth to them. Each company sets a certain price for a certain type of customer and then competes with other companies also going after the customer. The ad generally goes to the company that bids the highest. In other words, the value of showing their ads to that particular customer is the highest to them. The paper talks about a new algorithm that may be better at matching all the advertisers to potential customers. Exploits like pricing near the winning price in second-price auctions to exhaust competitors were also taken into account and a few safeguards were implemented.
In class, we’ve discussed auctions and the pricing strategies each player should pursue to maximize the potential gain from the auction. We also discussed the different types of auctions, with Adworks being a sealed English auction, where the price goes up. The good to be obtained here is the view of the customer. This specific algorithm for auctions is far more complicated than anything that was gone over in class, and takes into account the need to produce a competitive ratio, along with the runtime. The algorithm also heavily considers tradeoffs, which was not discussed much in the context of auctions for the class. In class, it was stated that for second-price auctions, the player should always bid their value as if they win, they will only pay less than their true value for the object. In an old version of the algorithm, it was stated that they had used an algorithm that used the highest value bid. Based on the lectures, this would have resulted in companies betting less than their true value in hopes of being able to save money, which would not maximize the profits for the advertising platform.