Psychotherapists in training lack a standardized and formalized method of patient interaction for proper development of empathy, communication, and experience. Currently, training involves residents practicing with each other, where one acts as the patient and one as the psychotherapist, or with simulated patients -actors who replicate patient scenarios. Both methods have shortcomings in availability, reliability, and the accuracy in replicating real scenarios. This project attempted to create virtual patients by utilizing online patient transcripts through the fine-tuning of three modern Artificial Intelligence models, ChatGPT-4o, LlaMa-3.1v-405B, and Gemini 1.5 Pro; as well as their miniature versions where applicable. A website interface was created to interact with the fine-tuned models for evaluation. The accuracy of the models was determined using cosine similarities to measure semantic relation between data and model outputs, ranging from 93.3% to 83.11% , with ChatGPT-4o Mini achieving the highest accuracy. These findings highlight the potential for virtual patients to serve as a more accessible, reliable, and effective training method for residents. Further evaluation and continual refinement remain necessary to address current limitations.
Abstract Psychotherapists in training lack a standardized and formalized method of patient interaction for proper development of empathy, communication, and experience. Currently, [...]
Depression is a global mental health issue generalized by a decrease in mood and satisfaction. Treatments for individuals afflicted with depressive symptoms include prescribed medications that require diagnosis to acquire. The purpose of this investigation was to accurately assist psychiatrists in diagnosis procedures to prevent both false positive and false negative conclusions by utilizing machine learning on social media messages. This was done by training a machine learning algorithm which accurately predicted and detected depressive behaviors and communications. As social media messages encompass individual’s general communications among long periods of time with high consistency and frequency, I hypothesized that social media messages could be used as a method to both train an accurate and consistent machine learning model for the detection of depression. Social media dataset messages rely on self-reported diagnoses. Based on F1 accuracy normalization across machine learning HyperTuning, average accuracy indicated 97% [+/-0.5%] among a ~7600 sample dataset. Utilizing generalized sentimental analysis has shown less accurate results (~80%) but needs further research.
Abstract Depression is a global mental health issue generalized by a decrease in mood and satisfaction. Treatments for individuals afflicted with depressive symptoms include prescribed [...]
Building a Conversational AI Medium to Enhance Psychotherapy Training with Virtual Patients 
