aiqbench analysis
This study critically evaluates the reliability of anxiety detection models by exposing how high keyword prevalence in training data inflates performance metrics, particularly for fine-tuned architectures. The findings suggest that frontier models exhibit less lexical dependence than smaller, domain-adapted counterparts, offering practitioners a viable path to robust inference without external API reliance. For developers building mental health tools, this highlights the necessity of evaluating models on text with specific diagnostic terms removed to assess true semantic understanding rather than pattern matching.
Reading Anxiety or Reading the Label? Comparing Fine-Tuned and Frontier Models for Anxiety Detection on Social Media
arXiv:2609.20847v1 Announce Type: new Abstract: Anxiety is among the most common mental health conditions, and people often write about it online well before seeking clinical help. Practitioners building detection tools face a concrete choice: call a frontier commercial model, fine-tune a smaller model in-house, or deploy a conventional classifier. We compare six conditions spanning all three on a held-out Reddit test set under a single controlled protocol. We also identify a confound in how this task is evaluated. In the corpus used here, 69.3% of anxiety-labelled posts contain the word "anxiety" or a variant, roughly twice the rate of comparable conditions, so a classifier can score well by keyword matching rather than by modelling the language of the condition. We therefore evaluate every model twice, on original text and with those terms deleted, and report the difference as lexical dependence. A frontier model leads on anxiety F1 (0.846), but a 110M-parameter domain-adapted encoder reaches 0.831 with no external API dependency, and mental-health domain pretraining accounts for only 0.7 of those points. Lexical dependence spans 8.6 to 25.4 points and does not track model capability: the LoRA fine-tuned 3B model is the most keyword-dependent condition tested, above even a TF-IDF classifier, while the frontier zero-shot model is the least. Published figures on this corpus are therefore upper bounds, and the inflation is largest for the fine-tuned models such figures typically report.
Title:Reading Anxiety or Reading the Label? Comparing Fine-Tuned and Frontier Models for Anxiety Detection on Social Media
Abstract:Anxiety is among the most common mental health conditions, and people often write about it online well before seeking clinical help. Practitioners building detection tools face a concrete choice: call a frontier commercial model, fine-tune a smaller model in-house, or deploy a conventional classifier. We compare six conditions spanning all three on a held-out Reddit test set under a single controlled protocol. We also identify a confound in how this task is evaluated. In the corpus used here, 69.3% of anxiety-labelled posts contain the word "anxiety" or a variant, roughly twice the rate of comparable conditions, so a classifier can score well by keyword matching rather than by modelling the language of the condition. We therefore evaluate every model twice, on original text and with those terms deleted, and report the difference as lexical dependence. A frontier model leads on anxiety F1 (0.846), but a 110M-parameter domain-adapted encoder reaches 0.831 with no external API dependency, and mental-health domain pretraining accounts for only 0.7 of those points. Lexical dependence spans 8.6 to 25.4 points and does not track model capability: the LoRA fine-tuned 3B model is the most keyword-dependent condition tested, above even a TF-IDF classifier, while the frontier zero-shot model is the least. Published figures on this corpus are therefore upper bounds, and the inflation is largest for the fine-tuned models such figures typically report.
Computation and Language (cs.CL); Social and Information Networks (cs.SI)
arXiv:2609.20847 [cs.CL]
(or arXiv:2609.20847v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.20847 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 340 words
