Ranking Model Adaptation For Domain-Specific Search

ABSTRACT:

With the explosive emergence of vertical search domains, applying the broad-

based ranking model directly to different domains is no longer desirable due to 

domain differences, while building a unique ranking model for each domain is 

both laborious for labeling data and time-consuming for training models. In this 

paper, we address these difficulties by proposing a regularization based 

algorithm called ranking adaptation SVM (RA-SVM), through which we can 

adapt an existing ranking model to a new domain, so that the amount of 

labeled data and the training cost is reduced while the performance is still 

guaranteed. Our algorithm only requires the Prediction from the existing 

ranking models, rather than their internal representations or the data from 

auxiliary domains. In addition, we assume that documents similar in the 

domain-specific feature space should have consistent rankings, and add some 

constraints to control the margin and slack variables of RA-SVM adaptively. 

Finally, ranking adaptability measurement is proposed to quantitatively 

estimate if an existing ranking model can be adapted to a new domain. 

Experiments performed over Letor and two large scale datasets crawled from 

a commercial search engine demonstrate the applicabilities of the proposed 

ranking adaptation algorithms and the ranking adaptability 

measurement.