Kaywan, Payam (2022) Human Depression Analysis: An Experimental Study of the Use of AI Botics for Early Detection. Research Master thesis, Victoria University.

Abstract

The world is facing a shortage of professional medical staff, a situation which has been exacerbated by the COVID-19 pandemic which has significantly increased challenges globally and has had an adverse impact on the health care system. This has also led to additional barriers to patient care access, specifically for individuals who are in need of constant special care. To address the issue of limited access to medical professionals, medical assistance can be provided to patients in the form of a chatbot which acts as a proxy between psychiatrists and patients and is available and accessible 24/7. Although there has been a degree of success in developing medical chatbots, many medical professionals believe that the use of chatbots in early depression detection needs to be more practical which will require further research. In this research, we address the well-known and common shortcomings which have been discussed in the recent literature. Three of these shortcomings are as follows: firstly, there is a lack of open-ended questions to enable participants to interact openly and without any restrictions about their moods and emotions as most bots in the literature constrain the participants’ responses by limiting them to multiple choice questions which means the participants are not able to open up and describe their real feelings freely. Secondly, there is a lack of semantic analysis to draw exact meaning from a text. Thirdly, there is a requirement for participants to make a long-term commitment in terms of their involvement in the research. This research introduces a depression analysis chatbot, DEPRA, which aims to resolve some of these shortcomings and challenges by asking open-ended questions, providing semantic analyses and automatic depression scoring. DEPRA is developed using contemporary bot platforms, Dialogflow on Google cloud-based infrastructure, and is integrated with social network platforms such as Facebook. Most chatbots today are designed for therapeutic purposes. However, the DEPRA chatbot is designed with a focus on the detection of depression in its early stages. DEPRA is designed based on a structured early detection depression Standard Interview Guideline the Hamilton Depression Scale (SIGH-D) and Inventory of Depressive Symptomatology (IDS-C), which is used by professional psychiatrists in triage sessions with patients. DEPRA has been trained with personalized utterances from a focus group. This research utilizes Natural Language Processing (NLP) to identify the depression level of participants based on their recorded conversation. DEPRA uses a scoring system to determine the participant’s depression level and severity. This research also details a non-clinical trial with 50 participants who interacted with the DEPRA chatbot. Due to the ethical limitations of this research, such as only residents of Australia and participants to be in the age group of 18 to 80 years old, we have approached a dataset with only 50 participants. This size of dataset was suitable to conduct and run the research. However, the future studies will target a more comprehensive dataset. This study was a first stage of utilizing Chatbot for early detection of depression. In this stage our goal was to develop the system not to run the clinical trial. Therefore, we required a sample that could assist us mainly to identify the accuracy of the system developed. Future work is to access evaluation by human expert which goes into the next phase of the project and could also include extending the sample and/or enhancing the system further and the assistance would be offered to western health. Therefore, at this stage 50+ sample sufficed to capture various responses by the people that had participants of different level of depression. To evaluate the autoscoring feature of DEPRA, the accuracy of the Machine Learning (ML) algorithms is calculated. Accordingly, manual scoring is compared with the calculated depression scores. The average accuracy of the 27 questions related to the linear SVC of the 26 participants’ experiment is 88%, the SGD algorithm of 40 participants’ experiment is 80%, and the linear SVC of 50 participants’ experiment is 87%. Furthermore, the overall satisfaction rate of using DEPRA was 79% indicating that the participants had a high rate of user satisfaction and engagement.

Search Google Scholar

Repository staff login