CHATBOT DEVELOPMENT IN INDONESIAN LANGUAGE TO HELP MENTAL HEALTH ANAMNESIS
COVID-19 pandemic also impacted people's mental health conditions and required treatment from experts. However, in Indonesia, there are only about 4150 psychologists and psychiatrists. Another impact of the pandemic is the restriction of face-to-face contact which opens up new challenges...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/56341 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | COVID-19 pandemic also impacted people's mental health conditions and required treatment
from experts. However, in Indonesia, there are only about 4150 psychologists and psychiatrists.
Another impact of the pandemic is the restriction of face-to-face contact which opens up new
challenges to start digitizing mental health services. An Indonesian-language chatbot was
developed as a solution for digitizing mental health services, especially for Anamnesis. Willie
et al. (2020) successfully developed IndoBERT as a state-of-the-art language model for the
Indonesian language that can be used in chatbot development. The purpose of this final project
is to design the architecture of the chatbot system, assess the addition of IndoBERT on the
chatbot model on performance, and assess the usability of the chatbot system.
The system was built on top of Rasa framework version 2.4. The system was made specifically
for the LINE platform. The system focused on three main features, mental health selfassessment, mental health advice, and the psychologist recommendation. The development
utilizes the line-bot-sdk to connect the server with LINE and the Google Places API for the
psychologist's recommendation. The system’s architecture is similar to the chatbot architecture
in general with the difference in the LINE connector and the usage of Google Places API. The
development is carried out using the waterfall methodology.
Experiments on adding IndoBERT showed a performance increase of about 0.2 on the F1-
Score of entity extraction and 0.01 on the intent classification when using IndoBERT-base.
However, there is a decrease in performance while other IndoBERT models are used. The
usability of the system was also tested with the help of 20 participants, which was carried out
virtually. All participants are students of the Bandung Institute of Technology. The usability
test results get a SUS score of 80 ± 12.22 which has passed the 68.0 benchmarks, a CUQ score
of 79.375 ± 9.67, and a UEQ that has passed the benchmark in all aspects tested. |
|---|