A0521
Title: A structural Mallows model for ranked data aggregation
Authors: Han Li - Shenzhen University (China) [presenting]
Abstract: The rank aggregation problem is studied, which aims to find a consensus ranking by aggregating multiple ranking lists. To tackle the problem probabilistically, an elaborate ranking model is formulated by generalizing the traditional Mallows model. The original model assumes a uniform pair preference structure, which imposes a strict condition on the data. The attempt is to relax this condition and propose a new model that allows the pair preference to vary structurally. The model is quite flexible and has a closed-form expression for complete rankings as well as top-k rankings. Several useful theoretical properties of the model are investigated, and efficient algorithms are proposed to infer the model structure and parameters. Through extensive simulation studies and real applications, the new model is demonstrated to have satisfactory performance in different scenarios.
