Lobbying the Narrative: Computational Analysis of Political Sentiment and Influence in U.S. Congress

Abstract

This thesis investigates the intersection of lobbying efforts and political sentiment in the United States Congress using a computational social science approach. Drawing on publicly available data from OpenSecrets.org, U.S. Congress, and congressional X (previously Twitter) posts, the study employs natural language processing (NLP), machine learning, and network analysis to examine how lobbying expenditures correspond to shifts in sentiment expressed by legislators across two major policy areas: health and taxation. Through sentiment analysis using VADER and RoBERTa models, topic modeling via LDA and BERTopic, and the construction of tripartite lobbying networks, the study identifies correlations between lobbying activity and changes in public-facing discourse, particularly among bill sponsors and in narratives surrounding salient policy debates. Moreover, the analysis finds that increased lobbying activity around a given issue often precedes an influx in the number of bills introduced on that topic, suggesting a connection between lobbying pressure and legislative agenda-setting. This research highlights the narrative power of interest groups in shaping both the rhetorical and procedural dynamics of policymaking, contributing to broader discussions about transparency, influence, and political communication in modern democracies.