Identifying Meaningful Frames Using Sentence-Level Embeddings & Network Analysis

• In Event: Eleventy Million Observations: Applications in Text Analysis & Machine Learning

Abstract

The study of framing has a rich methodological history, but to this point has mainly relied on time-intensive human coding schemes, both within qualitative and quantitative contexts. While advancements in machine learning have increased researchers’ ability to harness communicative texts for tasks such as topic classification and sentiment analysis, computational communication methods within the domain of framing have not kept pace. In this work we develop a methodology providing for the computational discovery of communication frames, using a combination of sentence-level embeddings and community detection in network analysis. We show the utility of the methodology using a bounded case study - debate over abortion in the leadup to repeal of the Republic of Ireland's 8th Amendment - demonstrating the evolution of framing within that debate and showing its applicability to more general studies of framing across broader issue domains.

Authors

• Mike Gruszczynski, Indiana University

• Kate Hunt, Indiana University-Bloomington