Papers by Anastassia Kornilova
An Item Response Theory Framework for Persuasion (2022.findings-naacl)
Copied to clipboard
| Challenge: | Several studies have considered the audience as a whole or by building separate models for different types of audiences. |
| Approach: | They propose a framework that can represent individual audience members in one model across a diverse set of persuasion tasks. |
| Outcome: | The proposed model performs well on three datasets including a novel dataset in the area of political advocacy. |
BillSum: A Corpus for Automatic Summarization of US Legislation (D19-54)
Copied to clipboard
| Challenge: | In the US Congress, over 10,000 bills are introduced each year, with state legislatures introducing tens of thousands of bills. |
| Approach: | They introduce the first dataset for summarizing US Congressional and California state bills . they demonstrate that models built on Congressional bills can be used to summarize California billa . |
| Outcome: | The proposed summarization methods can be applied to states without human-written summaries. |
Party Matters: Enhancing Legislative Embeddings with Author Attributes for Vote Prediction (P18-2)
Copied to clipboard
| Challenge: | Existing work on roll-call prediction limited to single session settings, thus not allowing for generalization across sessions. |
| Approach: | They propose a neural method that takes advantage of Congressional voting records to model voting behavior. |
| Outcome: | The proposed method achieves an average of 4% accuracy over the previous state-of-the-art. |
How Predictable is Your State? Leveraging Lexical and Contextual Information for Predicting Legislative Floor Action at the State Level (C18-1)
Copied to clipboard
| Challenge: | a study of state legislative initiatives shows that state legislatures have significant power over certain areas. |
| Approach: | They propose to use lexical content of over 1 million bills to build predictive models . they also use contextual legislature and legislator derived features to compare models based on state specific baselines . |
| Outcome: | The proposed models improve on baselines in all 50 states and D.C. lexical content, contextual features and legislative processes are used to build the models. |