FrameAxis: characterizing microframe bias and intensity with word embedding

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Social Sciences

Background:

• Framing influences perspectives by emphasizing specific aspects of an issue.
• Existing methods for analyzing framing often lack scalability or nuance.
• Computational approaches are needed to analyze framing in large text corpora.

Purpose of the Study:

• To introduce FrameAxis, an unsupervised method for characterizing documents based on semantic axes (microframes).
• To quantitatively measure "microframe bias" and "microframe intensity" within texts.
• To enable scalable and sophisticated computational analysis of framing effects across disciplines.

Main Methods:

• Utilizes word embeddings to identify overrepresented semantic axes (microframes) in text.
• Employs an unsupervised approach, eliminating the need for manual annotations.
• Quantifies microframe bias and intensity to characterize document framing.

Main Results:

• FrameAxis successfully identifies relevant semantic axes in diverse datasets (reviews, political news).
• Microframes with high bias and intensity correlate with sentiment, topic, and political leaning.
• The method provides nuanced insights into how framing is employed in text.

Conclusions:

• FrameAxis offers a scalable and nuanced computational method for analyzing framing.
• The approach can incorporate domain knowledge and explain results at word and document levels.
• This tool has the potential to accelerate framing research across various academic fields.