🏆 2024 Solution Challenge Top100 awarded
🏆 2024 AI 4 Good Hackathon Social Impact Prize
GDSC GIST
2024.1.16 ~ 2024.2.21
- Moodista is a mental health self-care app for people!
- Diary-based sentiment analysis 📝 provides you with customized contents and solutions!
- Support healthy life and well-being for people 🍀 by connecting them to counseling centers and treatment facilities!
- Moodista supports the UN SDG 3; Ensure healthy lives and promote well-being for all at all ages!
- Moodista makes mental health care more accessible; Break down the barriers to get traditional medical treatments!
- Moodista offers personalized mental health services; Make it easier for users to manage their mental health challenges!
This project is a starting point for a Flutter application.
A few resources to get you started if this is your first Flutter project:
For help getting started with Flutter development, view the online documentation, which offers tutorials, samples, guidance on mobile development, and a full API reference.
- install Flutter and Simulator
- play 'flutter pub get' in terminal to install environments
- Start debugging in 'main.dart' file
- flutter : 3.16.9
- dart : 3.2.6
- google maps flutter : 2.5.3
- google sign in : 6.2.1
- intl : 0.18.1
- firebase_core : 2.24.2
- firebase_auth : 4.16.0
- cloud_firestore : 4.14.0
app
├── assets
│ ├── fonts
│ └── images
├── lib
│ ├── constants
│ ├── models
│ ├── providers
│ ├── screens
│ ├── auth
│ ├── diary
│ ├── likes
│ ├── location
│ ├── recommended
│ └── settings
└── widgets
The user authentication system provided by firebase allows users to log in using email and Google login. The user's email information and the data to be searched based on the model's results were stored in the firestore. All the data that the user records on a daily basis is collected and managed using firestore.
To deploy the function and apply it to the app, we used Google cloud function. After using Bert Model to make it lightweight with tflite model, we imported the user's text from firestore and put it as the input of the function, and made it so that the app receives the result in json format. As a result, the app stores the information in firestore and uses it to recommend videos in the future.
- transformers : 4.35.2
- tensorflow : 2.15.0
- mediapipe-model-maker : 0.2.1.3
ai
├── dataset
│ ├── train.csv
│ └── validation.csv
├── model
│ └── mobilebert.tflite
└── mobilebert.py # Main code
Moodista employs MobileBERT, a lightweight version of the BERT model optimized for mobile devices, for text sentiment classification. We utilize MediaPipe Model Maker to customize the model, tailoring it to our specific data and applications.
The model is trained on emotion entries collected from the GoEmotions dataset, consisting of comments from Reddit users labeled with emotions like joy, sadness, fear, and anxiety. The dataset is preprocessed and formatted to suit the input requirements of the MobileBERT model.
The sentiment classification model is deployed on resource-constrained devices such as mobile phones or edge devices using TensorFlow Lite (TFLite) format, ensuring efficient execution with low latency and reduced memory footprint.
For more information on customizing models and datasets, refer to the online documentation and Kaggle page.
| Team Member | Role |
|---|---|
| Lee Jaehee | AI / Design |
| Ji Yuna | Front-end |
| Choi Wonhyeok | Back-end |
| Jeon Wooseok | AI |