BiliAnalyzer

A dashboard for reading a Bilibili creator's numbers: what they posted in a window of time, how it performed, what the audience said about it, and what an LLM makes of all three.

English · 繁體中文

Bilibili's creator centre shows totals and trend lines for your own account only, and nowhere on the site are comments or danmaku ever counted — they can only be read one video at a time. BiliAnalyzer answers the questions those pages cannot: what a creator's last few months actually looked like, which words and topics the audience responded to, whether the tone shifted, and how each audience segment behaved. It takes any public UID, so it works for a creator reviewing their own run, a fan or analyst studying someone else's, or anyone comparing two.

Give it a UID and a date range. It fetches every video the creator posted in that window, then charts the performance, tokenizes the titles, tags, subtitles, danmaku and comments into drillable word tables, scores sentiment, and crosses all of it against audience segments. An OpenAI-compatible endpoint can then be asked questions about the result, with tool calls against the same data.

Quick start

Requires Docker and Docker Compose.

git clone https://github.com/1morr/biliAnalyzer.git
cd biliAnalyzer
cp .env.example .env          # required — compose will not start without it
docker compose up --build -d

Then open http://localhost:8000. The API docs are at /docs on the same port; the built frontend is served by the same process, so there is no separate web port.

See it with data, immediately

Scraping a real creator takes a while and needs a SESSDATA cookie. To fill every panel with plausible data without one:

docker compose exec app python -m scripts.seed_demo

That seeds a fictional creator with a few months of videos, comments and danmaku. Pass --reset to start over.

Using it

1. New query — enter a creator's UID (the number in space.bilibili.com/546195) and pick a date range. The sidebar shows fetch progress.
2. Dashboard — summary tiles, view-count trend, interaction comparison, views-vs-engagement scatter, duration analysis, publish-hour density, and word tables you can click through to the underlying danmaku and comments. Audience segments (gender, membership, level, region) double as filter axes.
3. Video detail — one video's numbers with a radar chart and deltas against the query's own average.
4. AI analysis — configure an OpenAI-compatible Base URL, key and model in Settings, then ask follow-up questions; answers stream back.

SESSDATA is required for scraping. x/web-interface/view, which every video goes through, answers HTTP 412 to requests without a logged-in session — verified against the live API — so a run without it fails immediately rather than degrading. To add it: log in to bilibili.com, F12 → Application → Cookies → copy SESSDATA, paste it into Settings. Use the demo seed above to look around without one.

Security

This is a single-user local tool with no authentication. Every endpoint is open to whoever can reach the port, so Compose binds it to 127.0.0.1 only. Do not publish the port to a network you do not control, and do not put it behind a reverse proxy without adding auth first.

Your SESSDATA and API key are encrypted at rest with Fernet before they go into SQLite, and the API returns them masked. The encryption key lives in DATA_DIR/.secret_key — beside the database it protects, so this defends against someone getting the .db file alone, not against host access.

Tech stack

Layer
Frontend	Vite · React 19 · TypeScript · Tailwind CSS v4 · Base UI · ECharts
Backend	FastAPI · SQLAlchemy 2.0 (async) · aiosqlite · httpx
Chinese NLP	jieba (tokenizing) · SnowNLP (sentiment)
AI	OpenAI SDK, streamed over SSE, with tool calling
Deploy	Docker Compose, single multi-stage image

Configuration

Variable	Default
DATABASE_URL	sqlite+aiosqlite:///./data/bilianalyzer.db	Where the database lives
DATA_DIR	./data	Holds the database and .secret_key
SECRET_KEY	(generated)	Fernet key. Generated on first run and persisted; set it explicitly to keep stored credentials readable across environments
CORS_ORIGINS	http://localhost:5173	Only needed for the split dev setup. Empty disables CORS entirely

Local development

# backend
cd backend && python -m venv .venv && .venv\Scripts\Activate.ps1
pip install -e ".[dev]"
uvicorn app.main:app --port 8000 --reload

# frontend, in another terminal
cd frontend && npm install && npm run dev      # http://localhost:5173

pytest from backend/, npm run build and npx eslint . from frontend/. Requires Python 3.11+ and Node 20+.

License

MIT