SentiRank

SentiRank is a thesis-stage decision-support dashboard for Spotify review analysis. It combines IndoBERT sentiment inference, SVM aspect classification, and AHP/Fuzzy AHP research results behind a Next.js frontend and FastAPI microservices.

Active Architecture

Next.js frontend
-> API Gateway
-> review-service
-> sentiment-service (IndoBERT)
-> aspect-service (SVM)
-> decision-service (AHP/Fuzzy AHP)
-> report-service (Dashboard aggregation)
-> API Gateway repository persistence for runtime inference history

The frontend calls the API Gateway only. It does not call internal service ports, read CSV/JSON files directly, or calculate AHP/Fuzzy AHP. The standalone Reports page and print-report feature are out of the current thesis demo scope; report-service remains active as a Dashboard/evaluation/ranking aggregation service.

Local Demo

Start the default backend stack with SQLite runtime persistence:

docker compose up --build

Run the frontend locally:

cd frontend
npm install
npm run dev

Open http://localhost:3000. Set NEXT_PUBLIC_API_BASE_URL=http://localhost:8000 in the frontend environment.

Optional Compose modes:

# Include the frontend container
docker compose --profile frontend up --build
# Include PostgreSQL after setting API_GATEWAY_DATABASE_URL
docker compose --profile postgres up --build

Key server-side variables include API_GATEWAY_DATABASE_URL, DATABASE_URL, REVIEW_SERVICE_URL, SENTIMENT_SERVICE_URL, ASPECT_SERVICE_URL, DECISION_SERVICE_URL, REPORT_SERVICE_URL, SENTIMENT_MODEL_SOURCE, INDOBERT_MODEL_PATH, INDOBERT_MODEL_ID, HF_TOKEN, and SVM_ASPECT_MODEL_PATH. Use .env.example and frontend/.env.example as templates; never commit real secrets.

Data and Models

• Research CSV/JSON artifacts remain reproducible, read-only thesis outputs for datasets, preprocessing, evaluation, AHP/Fuzzy AHP, ranking comparison, and Dashboard snapshots.
• Runtime inference history stores user-submitted text, predictions, model/provenance metadata, and timestamps through the API Gateway repository.
• SQLite is the local/demo default; PostgreSQL is optional for deployment.
• Legacy Prisma files have been removed and Prisma is not part of runtime persistence.
• Local artifacts under ml-service/saved_models/ are ignored by Git. Services can use read-only local mounts; sentiment serving also supports a configured Hugging Face model id.

Deployment Modes

Mode	Frontend	Backend	Database	Model source
Local full demo	Local npm run dev or Compose frontend profile	Docker Compose default stack	SQLite	Read-only local artifacts or Hugging Face id
Semi-online demo	Vercel	Local Docker backend through a tunnel	Local SQLite	Local backend artifacts or Hugging Face id
Full online, if required	Vercel	Container-capable platform	Managed PostgreSQL	Hugging Face/private model repository or platform-managed artifact

Vercel hosts the frontend only; it does not run the full Docker Compose backend.

Repository Roles

Path	Role
frontend/	Next.js UI and API Gateway client layer.
services/	Active API Gateway and domain runtime microservices.
ml-service/	Research scripts, notebooks, quality audits, model utilities, and legacy experiment runtime code; not the primary runtime service layer.
datasets/	Research datasets and generated research outputs.
docs/	Thesis, methodology, frontend, and architecture documentation.
ml-service/saved_models/	Local Git-ignored model artifacts.

See docs/microservices/architecture.md for canonical service ownership and docs/microservices/docker-compose-foundation.md for detailed runtime commands.