Automatic

Paid search advertising has been a major form of online advertising in recent years. In this form of advertising, an advertiser submits a list of keywords to major search engines. When one of the keywords matches the query keyword that a search engines user submits, the ad of this advertiser will have a chance to be shown on the search result page. If the user is interested and clicks on the ad, the advertiser will be billed of each clickthrough with a predetermined cost-per-click fee by the search engine, regardless whether the user purchases anything after entering the advertiser’s website. The advertiser will try to make a profit by hoping a higher probability that a clickthrough can end with a sale. So in an ad campaign the fundamental question the advertiser wants to ask is: what are the good keywords that can attract more clickthrough traffic form search engines, and more importantly have higher sale conversion rate given these clickthrough traffic? Several marketing literatures have addressed this issue but their keywords selection and evaluation methods need human interactions. Our goal is to try to develop a statistical learning method that can automate the keyword evaluation processes. This paper is our pilot study and we want to know whether such statistical learning method can really the same or even better job than human As a comparison, we compared our result with another study with the same data but using mainly manual evaluation process. The result shows our method has better prediction accuracy.