I’m a clinical psychologist who worked with psychometric scale development and some NLP (mostly topic models for my PhD). I’ve worked through fast.ai 2019 courses and I’m currently reading the fast.ai book.
I’m interested in current applications and project ideas for applying DL in psychology or, if possible, in clinical psychology. I’ve seen some projects, but they seem mostly untenable or underwhelming:
- Chatbots as a proxy for counseling services: DL language models might help make more interactive chatbots, but counseling in general is too much artisanal for a bot to really work effectively beyond simple advice;
- Diagnose or identify at-risk persons from unstructured data: this one is really interesting, but there are some problems: mental disorders nosology is quite messy (with lots of overlap and ambiguity) and reality is complex - comorbidities are common, e.g. So, building a labeled dataset is not a simple task (the labels are guaranteed to be noisy, at the very least); but even if this was not a problem, most data we have readily available are not labeled.
- Use RNNs to predict disease progression: most parametric models currently in use to understand intensive time series data obtained from ecological momentary assessment experiments (in which a single subject is assessed multiple times a day, during a long period, on some set of symptoms or a psychometric construct of interest) are known to be problematic in their assumptions; using RNNs might be a way to model those time series in a more flexible way. But deriving theoretical insights from NN is not simple nor always tenable (can we really better understand language by exploring the hidden states of a giant language model?).
If anyone is aware of projects that are successfully applying DL in psychology, please share!