Auto-scaled Bayesian browsing model in massive data

Liyun Ru, Anhui Wang, Yingying Wu, Shaoping Ma

Abstract: In the field of information retrieval, the method of building a click model by mining click logs to improve the effect of the search engine has been widely studied. And Bayesian Browsing Model (BBM), for the calculation of the operability and the effectiveness of the result, is widely used. However, when applied in engineering, especially for the situation of large scale data, this model will not perform properly. This problem is described analytically and shown by numerical experiments in this paper. For this problem, an auto-scaled BBM method is proposed. Experiments show that the new method solves the problem of original BBM, and have a better performance in terms of NDCG.