Roll Call Vote Prediction with Knowledge Augmented Models

Proceedings of the 23rd Conference on Computational Natural Language Learning, pages 574–581 Hong Kong, China, November 3-4, 2019. c⃝2019 Association for Computational Linguistics 574 Roll Call Vote Prediction with Knowledge Augmented Models Pallavi Patil♦ Kriti Myer♦ Ronak Zala♦ Arpit Singh♦ Sheshera Mysore♦ Andrew McCallum♦ Adrian Benton♠ Amanda Stent♠ ♦College of Information and Computer Sciences University of Massachusetts Amherst ♠Bloomberg LP {ppatil,kmyer,rzala,arpitsingh,smysore,mccallum}@cs.umass.edu {abenton10,astent}@bloomberg.net Abstract The official voting records of United States congresspeople are preserved as roll call votes. Prediction of voting behavior of politicians for whom no voting record exists, such as individ- uals running for office, is important for fore- casting key political decisions. Prior work has relied on past votes cast to predict future votes, and thus fails to predict voting patterns for politicians without voting records. We ad- dress this by augmenting a prior state of the art model with multiple sources of external knowledge so as to enable prediction on un- seen politicians. The sources of knowledge we use are news text and Freebase, a man- ually curated knowledge base. We propose augmentations based on unigram features for news text, and a knowledge base embedding method followed by a neural network com- position for relations from Freebase. Em- pirical evaluation of these approaches indi- cate that the proposed models outperform the prior system for politicians with complete his- torical voting records by 1.0% point of ac- curacy (8.7% error reduction) and for politi- cians without voting records by 33.4% points of accuracy (66.7% error reduction). We also show that the knowledge base augmented ap- proach outperforms the news text augmented approach by 4.2% points of accuracy. 1 Introduction Roll call votes are official records of how politi- cians vote on bills (potential laws) in the United States House of Representatives and Senate. Reli- able prediction of these votes, using historical vot- ing records and the text of bills, can be used to forecast political decisions on key issues, which can be informative for the electorate and other po- litical stakeholders. Prior work has used politi- cians’ voting records as a means to study their ide- ological stances (Poole and Rosenthal, 1985; Clin- ton et al., 2004), as well as roll call votes com- bined with the text of the corresponding bills to predict votes on newly drafted bills (Gerrish and Blei, 2012; Kraft et al., 2016; Kornilova et al., 2018). However, these approaches fail to make good predictions for the votes of politicians whose records are not established, such as new candidates for office – a time when this information can be most useful for the electorate. We hypothesize that additional sources of knowledge about new politi- cians can predict their future votes. In this work we explore two sources of addi- tional knowledge about politicians to better pre- dict roll call votes: news article text about the politicians, and Freebase (Bollacker et al., 2008), which is a manually curated knowledge base (KB). Relevant news articles may contain words that are indicative of a politician’s stance with …

Content truncated to 500 words