Farhad Mohsin

Inwon Kang, Sikai Ruan, Tyler Ho et al.

Finding preferences expressed in natural language is an important but challenging task. State-of-the-art(SotA) methods leverage transformer-based models such as BERT, RoBERTa, etc. and graph neural architectures such as graph attention networks. Since Large Language Models (LLMs) are equipped to deal with larger context lengths and have much larger model sizes than the transformer-based model, we investigate their ability to classify comparative text directly. This work aims to serve as a first step towards using LLMs for the CPC task. We design and conduct a set of experiments that format the classification task into an input prompt for the LLM and a methodology to get a fixed-format response that can be automatically evaluated. Comparing performances with existing methods, we see that pre-trained LLMs are able to outperform the previous SotA models with no fine-tuning involved. Our results show that the LLMs can consistently outperform the SotA when the target text is large -- i.e. composed of multiple sentences --, and are still comparable to the SotA performance in shorter text. We also find that few-shot learning yields better performance than zero-shot learning.

Farhad Mohsin

The group no-show paradox (GNSP) occurs when a group of agents abstaining from voting can make the new winner more preferred to them. Previous work has suggested that even for voting rules susceptible to this paradox, it is a rare occurrence in real elections and under various assumptions. However, we find that under one-dimensional preference models such as 1D-Euclidean, single-peaked, or single-crossing preferences, Single Transferable Vote (STV), a popular runoff rule, is highly vulnerable to GNSP. This is in stark contrast to Condorcet rules, another family of rules susceptible to GNSP, where the paradox cannot occur under these one-dimensional preferences. We theoretically identify tractable and prevalent sufficient conditions for GNSP to occur for STV under one-dimensional preference models. Through our theoretical results and experiments with synthetic preference profiles from these domains, we demonstrate that voters at the extremes of the 1D spectrum are particularly likely to cause GNSP by abstaining. Furthermore, the likelihood of occurrence increases substantially as the number of alternatives grows.