POLITICS: Pretraining with Same-story Article Comparison for Ideology Prediction and Stance Detection

Abstract

Ideology is at the core of political science research. Yet, there still does not exist generalpurpose tools to characterize and predict ideology across different genres of text. To this end, we study Pretrained Language Models using novel ideology-driven pretraining objectives that rely on the comparison of articles on the same story written by media of different ideologies. We further collect a large-scale dataset, consisting of more than 3.6M political news articles, for pretraining. Our model POLITICS outperforms strong baselines and the previous state-of-the-art models on ideology prediction and stance detection tasks. Further analyses show that POLITICS is especially good at understanding long or formally written texts, and is also robust in few-shot learning scenarios.

Analyzing the Stance of Facebook Posts on Abortion Considering State-Level Health and Social Compositions

“…However, recently, several studies approach stance detection by using BERT-based models (Kawintiranon and Singh 2021;Liu et al 2021;Alturayeif, Luqman, and Ahmed 2022;Glandt et al 2021;Clark et al 2021;Barbieri et al 2020) reporting average F1 scores in range approximately between 0.7 to 0.9. Prior literature found that BERT-based models outperform other models in stance detection on SemEval 2016 dataset (Ghosh et al 2019) reaching state-of-the-art performance with accuracies close to or above 0.9 (Slovikovskaya 2019;Dulhanty et al 2019;Liu et al 2022).…”

“…To detect the stance of Facebook posts about abortion, we developed a model using a transfer learning approach, using XLNet (Yang et al 2019) and RoBERTa (Liu et al 2019) and fine-tuning them on a ground truth dataset. While deep learning models have been used for stance detection on other topics, such as political debates (Kawintiranon and Singh 2021;Liu et al 2022), or COVID-19 issues (Glandt et al 2021), to the best of our knowledge, this work is the first to develop an abortion stance detection model.…”

“…Both are large language models based on the Transformer architecture (Vaswani et al 2017), and both models regularly achieve very high accuracy on many prediction tasks. While there are studies employing XLNet (SU et al 2021;Yang et al 2019), previous research suggests that RoBERTa can achieve state-of-the-art results outperforming BERT and XLNet (Slovikovskaya 2019;Dulhanty et al 2019;Liu et al 2022;Barbieri et al 2020). Despite its strong performance, RoBERTa has a maximum input length when processing texts.…”

ArguSense: Argument-Centric Analysis of Online Discourse

“…Within the field of argument and stance detection there are several works utilizing NLP to classify different niche grammatical devices. Liu et al (2022) uses pretrained language models to characterize and predict ideologies across different genres of text. Zhang et al (2022) uses graph neural networks to classify political perspectives.…”

Trusting the Moral Judgments of a Robot: Perceived Moral Competence and Humanlikeness of a GPT-3 Enabled AI

“…The toxicity measure is calculated by HateBert (Caselli et al, 2021;Liu et al, 2022), a machine-learning algorithm that Delphi outputs queries to. Toxicity Score is a measure of how offensive a query is and ranges from 0.00 -1.00 with higher scores representing more offensive queries.…”