Extracting Consumer Insight from Text: A Large Language Model Approach to Emotion and Evaluation Measurement
A new fine-tuned LLM outperforms GPT-4 Turbo and achieves over 95% accuracy on Amazon and Yelp reviews.
Deep Dive
A research team led by Stephan Ludwig introduces the Linguistic eXtractor (LX), a fine-tuned large language model for measuring consumer emotions and evaluations from text. LX achieves 81% macro-F1 accuracy on survey responses and over 95% on annotated reviews, outperforming GPT-4 Turbo and RoBERTa. It identifies 16 emotions and 4 evaluation constructs, proving emotions like discontent directly influence purchase behavior. A free, no-code web app is available for scalable analysis.
Why It Matters
Enables brands to extract precise emotional signals from reviews at scale, moving beyond simple star ratings to predict consumer behavior.