Preference learning has found applications in many areas, such as meta-search, information retrieval, and recommender systems. Algorithms to learn preference from linear orders (full rankings) have been well-investigated. However, many real-world datasets are composed of partial orders over a subset of alternatives (e.g., Netflex data). Two key questions in preference learning are of great interests.
First, if partial orders are the only information given, can we provide good measure to the similarity between agents? Second, can we effectively learn preference without ignoring any information in the dataset?
In this talk, the speaker will give affirmative answer to both questions. In the first part of this talk, we will focus on the inefficacy of the widely-used Kendall-tau distance and propose a theoretically guaranteed measure to similarities between agents. Second part of this talk present an effective preference learning algorithm without ignoring any information. Practical performance of both parts will be also demonstrated together with theoretical analysis.
2019-08-13 14:00 ~ 15:00
Ao Liu, PhD student in Rensselaer Polytechnic Institute
Room 602,School of Information Management & Engineering, Shanghai University of Finance & Economics