Directed Social Regard: A New NLP Approach for Multi-Valence Sentiment Analysis
A new paper on arXiv (2605.00776) introduces Directed Social Regard (DSR), a multi-dimensional sentiment analysis method that detects both positive and negative sentiments in online messages and identifies their targets. Unlike traditional NLP tools that classify overall sentiment as positive, neutral, or negative, DSR uses transformer-based models to (1) identify span-level targets of sentiment and (2) score each span along three axes of regard (ranging from -1 to 1), grounded in social science theories of moral disengagement and moral framing. The approach addresses the coexistence of pro-social and anti-social sentiments in the same message, common in online platforms, influence operations, and political rhetoric. The paper is a preprint and has not yet been peer-reviewed.
Key facts
- The paper is published on arXiv with ID 2605.00776.
- DSR stands for Directed Social Regard.
- DSR is a multi-dimensional, multi-valence sentiment analysis approach.
- It uses a pair of transformer-based models.
- One model detects span-level targets of sentiment.
- The other model scores spans along three axes of regard.
- The axes are motivated by moral disengagement and moral framing theories.
- Traditional NLP tools classify overall sentiment as positive, neutral, or negative.
Entities
Institutions
- arXiv