Adapting Word Embeddings from Multiple Domains to Symptom Recognition from Psychiatric Notes

Area of Science:

• Natural Language Processing
• Computational Psychiatry
• Medical Informatics

Background:

• Accurate identification of psychiatric symptoms is crucial for mental health diagnosis and treatment personalization.
• Conventional natural language processing (NLP) methods struggle with the complexity and variability of symptom data in clinical text.
• Developing advanced NLP techniques is essential to overcome these challenges in mental healthcare.

Purpose of the Study:

• To adapt and evaluate word embeddings from diverse external domains for enhanced psychiatric symptom recognition in clinical text.
• To investigate the effectiveness of different word embedding adaptation strategies for improving NLP in psychiatry.
• To establish a novel approach for leveraging external knowledge to improve automated analysis of mental health data.

Main Methods:

• Adapted word embeddings from four source domains: intensive care, biomedical literature, Wikipedia, and Psychiatric Forum.
• Investigated four adaptation methods: using source embeddings directly, combining source and target data, weighted embeddings, and retraining with target domain data.
• Evaluated the performance of these methods for psychiatric symptom recognition in clinical text.

Main Results:

• The strategies involving weighted word embeddings and retraining source models with target domain data significantly outperformed baseline methods.
• These findings demonstrate the efficacy of adapting external word embeddings for psychiatric symptom extraction.
• This study represents the first attempt to adapt multiple external word embeddings for improved psychiatric symptom recognition.

Conclusions:

• Adapting word embeddings from external domains, particularly through retraining with target data, enhances psychiatric symptom recognition.
• The proposed methods offer a promising solution for overcoming NLP limitations in analyzing complex mental health clinical text.
• This work paves the way for more accurate and personalized mental healthcare through advanced computational approaches.