USI - Email

Archive / INF Seminars / INF_2025_04_15_LuizaPiancastelli

USI - Email

	Recent Advances in Statistical Models for Rankings

Host: Prof. Antonietta Mira

Martedì

15.04

USI Campus EST, Room D1.14

12:30 - 13:30


	Luiza Piancastelli University College Dublin

You can join online by clicking here.

Abstract: This talk introduces distance-based probabilistic models for rankings and goes over two recent advancements in this area. Ranking data refers to information that has been arranged in a specific order. It is increasingly abundant in recommendation systems and is natural in sports and elections. However, rankings require special attention due to their specific multivariate and relative nature. In the first part of the talk, we deal with the problem of ranking aggregation - preferences from multiple evaluators are observed and the goal is to summarise their opinions. The Mallows model (Mallows, 1957) is one of the most famous approaches for this, summarising the data in terms of a modal rank, the consensus. But what if the judge body is not decisive about all ranking positions? For example, there might be strong evidence of which are the best and worst options, but less so on the middle ranks. The Clustered Mallows Model (CMM, Piancastelli and Friel 2025) is an extension that accommodates indifference in the consensus by clustering options. In the second part of the talk, our focus is on rankings that change in time. Motivated by the weekly rankings of tennis players, we ask: how can we model ranking dynamics and carry out predictions? To this aim, we propose the R-GARCH model (Piancastelli and Barreto-Souza ArXiv) that combines the Mallows distribution with an autoregressive-moving average type structure. Model strengths include an elegant mathematical formulation, clear conditions for stationarity and ergodicity, easy parameter interpretation, and a solid strategy for handling incomplete data.

Biography: Dr Luiza Piancastelli is a lecturer at University College Dublin. Her research focuses on ranking data, time series analysis, clustering, and regression models, with a particular emphasis on Bayesian inference. She is passionate about developing statistical methodologies that address complex data structures.