Narratives are accounts of relationships between entities such as people, places, concepts, or facts that influence how individuals understand and mentally represent reality. Political Science has long recognized the importance of narratives as drivers of major political change. A limitation to the study of narratives in the context of social media is that narratives require the quantification of who is doing what and to whom, which necessitates recovering lower-dimensional narrative representations of plain text. This project uses Semantic Role Labeling (SRL) to distill complex narrative elements into lower-dimensional forms. It uses SRL on Colombian politicians' tweets to extract political narratives and assess narrative shifts following the 2022 electoral cycle.

Tweets were scraped with the GetOldTweets-Python project, translated with a pre-trained Hugging Face transformer and SRL was performed with the relatio Python package with functions from spaCY, Gensim, and pre-trained Glove embeddings. Narratives are represented on a directed graph with NetworkX.