Improving click model by combining mouse movements with click-through data

As a typical user activity based model in search engines, click model is currently the mainstream approach to describe and analyze user behaviors when handling a number of search-related applications, such as automated ranking alternations, search quality metrics, and online advertising. However, most of the existing work in this literature only consider click through data but ignore other aspects of the interactions between users and search results (e.g. eye and mouse movements, etc.). To predict the click-through rates (CTRs)
¹
more accurately, this paper improves the click model by combining mouse tracking with click through data to obtain more objective measures of user experience and consequently achieve a better understanding of user behaviors. Experimental results that are uniformly sampled from the most popular Chinese search engine - Baidu.com
²
show that the proposed approach outperforms the existing models at predicting CTRs. Online test results covering 25% of the daily Internet traffic in Baidu.com show that the top-part ranking list adjusted based on the predicted CTRs increase the CTR of the first URL and the total CTRs of the URLs in the first page. These experiments prove that the proposed approach improves the quality of search results and meets users' information needs more accurately.