Automatic depression severity assessment with deep learning using parameter-efficient tuning.

Abstract

To assist mental health care providers with the assessment of depression, research to develop a standardized, accessible, and non-invasive technique has garnered considerable attention. Our study focuses on the application of deep learning models for automatic assessment of depression severity based on clinical interview transcriptions. Despite the recent success of deep learning, the lack of large-scale high-quality datasets is a major performance bottleneck for many mental health applications. A novel approach is proposed to address the data scarcity problem for depression assessment. It leverages both pretrained large language models and parameter-efficient tuning techniques. The approach is built upon adapting a small set of tunable parameters, known as prefix vectors, to guide a pretrained model towards predicting the Patient Health Questionnaire (PHQ)-8 score of a person. Experiments were conducted on the Distress Analysis Interview Corpus - Wizard of Oz (DAIC-WOZ) benchmark dataset with 189 subjects, partitioned into training, development, and test sets. Model learning was done on the training set. Prediction performance mean and standard deviation of each model, with five randomly-initialized runs, were reported on the development set. Finally, optimized models were evaluated on the test set. The proposed model with prefix vectors outperformed all previously published methods, including models which utilized multiple types of data modalities, and achieved the best reported performance on the test set of DAIC-WOZ with a root mean square error of 4.67 and a mean absolute error of 3.80 on the PHQ-8 scale. Compared to conventionally fine-tuned baseline models, prefix-enhanced models were less prone to overfitting by using far fewer training parameters (<6% relatively). While transfer learning through pretrained large language models can provide a good starting point for downstream learning, prefix vectors can further adapt the pretrained models effectively to the depression assessment task by only adjusting a small number of parameters. The improvement is in part due to the fine-grain flexibility of prefix vector size in adjusting the model's learning capacity. Our results provide evidence that prefix-tuning can be a useful approach in developing tools for automatic depression assessment.