UPPAM: A Unified Pre-training Architecture for Political Actor Modeling based on Language

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 11996–12012 July 9-14, 2023 ©2023 Association for Computational Linguistics UPPAM: A Unified Pre-training Architecture for Political Actor Modeling based on Language Xinyi Mou1, Zhongyu Wei1,2∗, Qi Zhang3, Xuanjing Huang3,4 1School of Data Science, Fudan University, China 2Research Institute of Intelligent and Complex Systems, Fudan University, China 3School of Computer Science, Fudan University, China 4Shanghai Collaborative Innovation Center of Intelligent Visual Computing, China {xymou20,zywei,qz,xjhuang}@fudan.edu.cn Abstract Modeling political actors is at the core of quan- titative political science. Existing works have incorporated contextual information to better learn the representation of political actors for specific tasks through graph models. However, they are limited to the structure and objective of training settings and can not be general- ized to all politicians and other tasks. In this paper, we propose a Unified Pre-training Ar- chitecture for Political Actor Modeling based on language (UPPAM). In UPPAM, we aggre- gate statements to represent political actors and learn the mapping from languages to represen- tation, instead of learning the representation of particular persons. We further design structure- aware contrastive learning and behavior-driven contrastive learning tasks, to inject multidimen- sional information in the political context into the mapping. In this framework, we can pro- file political actors from different aspects and solve various downstream tasks. Experimen- tal results demonstrate the effectiveness and capability of generalization of our method. 1 Introduction Political actors are shaping our attitudes, opinions, and decisions toward public issues. For instance, on social platforms, politicians can select and em- phasize certain aspects of content to bias the dis- cussion, through which they can derive an opinion climate from user engagement and acquire direct feedback from potential voters and opinion lead- ers (Bene, 2017; Heiss et al., 2019). Political actor modeling is essential for quantitative political sci- ence and has applications in various downstream tasks such as roll call vote prediction (Yang et al., 2020), frame detection (Johnson et al., 2017) and bias detection (Baly et al., 2020). Data-driven approaches utilize different kinds of information to profile political actors, including public statements, legislative behaviors and social ∗Corresponding author. • No church needs to provide contraception under ObamaCare. • Recovery package must provide state aid, hazard pay. • LGBT rights are in jeopardy from Supreme Court. • I oppose school busing because it fails, not for racism. languages social network behaviors Voted NO on defining unborn child as eligible for SCHIP. Voted NO on constitutional ban of same-sex marriage. Voted YES on Educational Savings Accounts. Joe Biden Figure 1: An illustration of political actors. They not only participate in legislative activities, but also form relationships with others, and convey opinions through tweets, speeches and etc. We propose to represent po- litical actors based on their statements and learn the mapping from language to their representations using social networks and behaviors as self-constructed super- vision. networks (Figure 1). Early research analyzes roll call data to estimate the ideology of political ac- tors. …

Content truncated to 500 words