Prediction of keyword auction using Bayesian Network

Online keyword auctions, in which marketers bid for advertising slots along the search engine results, have become a new channel of advertisement. To better manage the advertisement campaign, a key challenge for advertisers is to predict each keyword's bidding price and effectiveness (e.g. click through rate), which are not priorly known to the individual advertiser. This paper identifies those relevant variables affecting auction strategy and models them in causal connections using history data in order to simulate the bidding behavior. We verified the effective necessaries of these predictions using empirical auction data, and our result indicated that the prediction with Bayesian Network produce close-to-reality results.