Thank you @veerleceline.bsky.social, @alexgu85.bsky.social,
@katarinakjell.bsky.social, @salgiorgi.bsky.social, Roman Kotov, @adigan.bsky.social, @handyschwartz.bsky.social and @oscarkjell.bsky.social for great collaboration 🙏
Many of the resulting models converge with rating scales at r > .80 — approaching the outcomes’ own reliability, the theoretical ceiling for concurrent accuracy.
And applying a model is one line of R:
textAssess(model_info = "depression_text_phq9_roberta23_gu2024", texts = “I hate the replication crisis”)
Language in, score out. 🎯
The library is an achievement of interdisciplinary work between psychology and computer science.
We hope researchers will enjoy the library, both as users and contributors.
Over the years, we have built many many models where we take rich language as input (e.g., “describe how you are feeling”) and train models to predict an outcome (e.g., self-rated emotions), using LLMs and machine learning.
So why a library of such models? Because L-BAMs have mostly been one-off artefacts attached to papers — hard to find, hard to reuse, hard to validate independently. The L-BAM library gives them standardised documentation: training data, performance metrics, generalisability info etc.
Anyone can contribute models — there's a documentation template covering outcome, training data, validation, and ethics. The goal is a shared resource the field maintains together, not a one-lab repo. 🤝
docs.google.com/spreadsheets...
New AMPPS paper introducing a new LIBRARY: the Language-Based Assessment Model (L-BAM) Library 📚 with 50+ open models for assessing mental health, well-being, implicit motives, and more 🧵
journals.sagepub.com/doi/full/10....
1/5 Did you ever wish to make a visualization of language data? 🤔
We made an R-based tutorial for social scientists on how to turn language data into visual insights âś…
Preprint: doi.org/10.31234/osf...
