@inproceedings{pandey-etal-2026-automatic,
    title = "Automatic Annotation of Mental Health Recovery Narratives: A Benchmark Study",
    author = "Pandey, Shrankhla  and
      Murray, Graham  and
      Laws, Ben  and
      Rennick-Egglestone, Stefan  and
      Slade, Mike  and
      Morgan, Sarah",
    editor = "Zirikly, Aya  and
      Bar, Kfir  and
      MacAvaney, Sean  and
      Ireland, Molly  and
      Ophir, Yaakov  and
      Atzil-Slonim, Dana  and
      Varadarajan, Vasudha  and
      Bedrick, Steven  and
      Desmet, Bart",
    booktitle = "Proceedings of the 10th Workshop on Computational Linguistics and Clinical Psychology ({CLP}sych 2026)",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.clpsych-1.8/",
    pages = "100--118",
    ISBN = "979-8-89176-421-7",
    abstract = "Manual annotation of mental health recovery narratives is slow and emotionally demanding, which limits the scalability of the digital mental health resource. A framework exists to characterise such narratives, called INCRESE, but there are currently no methods to automatically annotate the characteristics defined in INCRESE. We benchmarked the ability of support vector classifiers to annotate INCRESE characteristics when trained with three families of text representations: bag of words, GloVe static embeddings, and BERT contextual embeddings, using a dataset of 355 mental health recovery narratives. Characteristics related to diagnosis and turning points achieved a balanced accuracy greater than 0.67. Characteristics related to content warnings achieved a balanced accuracy of 0.72 but showed poor recall, which may be harmful for readers because it could lead to unsolicited exposure to sensitive content such as abuse or sexual violence. The lived-experience advisors endorsed the project objectives and addressed challenges of characteristic prioritization, adding insights not visible from quantitative metrics alone."
}