Linked Latent Theta Roles: A Model to Support Social Scientists with Open-ended Exploration of Framing

ACL ARR 2025 May Submission472 Authors

Abstract: Computational research has developed techniques to classify frames in text. However, these techniques may be less useful for supporting researchers in exploratory analysis of framing as an act of meaning construction. To address this gap, we introduce Latent Linked Theta Roles (LLTR), a model based on linguistic attributes relevant to framing language. Rather than identifying frames per se, the LLTR model highlights linguistic patterns that might be indicative of framing, thus supporting researchers in conducting open-ended, exploration of framing. A qualitative human-subject study compares this novel model against two baseline models, demonstrating that LLTR is more effective in assisting researchers with this exploratory task.

Paper Type: Long

Research Area: Human-Centered NLP

Research Area Keywords: Framing, topic modeling, qualitative evaluation

Contribution Types: Model analysis & interpretability

Languages Studied: English

Keywords: social media

Submission Number: 472