Research On On-line Advertising Algorithm

On-line advertising springs up as a new advertising industry with the rapid development and popularity of network. The model of on-line advertising is also changing. In this paper, we focus on two main advertising models:sponsored search advertising and content-targeted advertising.Search engine advertising has become a significant element of the Web browsing experience. Choosing the right advertisements for the query and the order in which they are displayed greatly affects the probability that a user will see and click on each advertisement. This ranking has a strong impact on the revenue the search engine receives from the advertisement. Further, showing the user an advertisement that they prefer to click on improves user satisfaction. For these reasons, it is important to be able to accurately estimate the click through rate of advertisements in the system.The following is my work:1. Research on prediction of click through rate on sponsored search advertising. When we calculate each data’s label, its click through rate is not only related to advertisement’s content, but also to the page number of advertisement and the location of its page. Considering the above factors, we use maximum likelihood estimation method to get each advertisement’s label in training data. We use three feature selection methods, information gain, chi-square and RReliefF algorithm. Also, we use three regression algorithms:logistic regression model, support vector machine and gradient boost decision tree. And these algorithms are compared with each other and then to be analyzed. The experiments show that RReliefF algorithm outperforms the other two algorithms, and both gradient boost decision tree and support vector machine have good performances. Of the two algorithms, gradient boost decision tree performs better.2. We propose a new algorithm based on the characteristic of content-targeted advertisement, which has been improved on BM25algorithm. Word expansion is added to the improved BM25algorithm to reduce the length of text and develop the matching accuracy. Experiments show that the improved BM25algorithm solves mismatch between web page texts and advertisement texts.In summary, for prediction of click through rate on sponsored search advertising, we experiment on different feature selection methods and regression algorithms with the changing of data sets and feature numbers. And get the meaningful conclusion. For the algorithm of content-targeted advertising, the new proposed algorithm improves matching quality between the web page and the advertisement.