bert-sentiment-tfjs

Sentiment Analysis using BERT model and Tensorflowjs

Instruction:

1. Run npm install first to install the required modules.
2. Run npm run dev to run to start the server.

Run analysis directly on browser (GPU)

Open http://localhost:3000 in browser. Open web console in browser to see output which should look like:

Loaded Tokenizer.
105 Model Loading time (ms): 4941

Run analysis from the server side

Open http://localhost:3000/server.html in your browser.

Convert vocab.txt to vocab.json (optional)

• An included vocabulary file, vocab.json, is extracted and converted from "Bert-base, uncase". If you would like to convert other pre-trained vocabularies use the following example:

• cd src/util
  python txt2json.py ../../public/vocab.txt /tmp/vocab.json
  

Run Debugger for Chrome in vscode (optional)

1. Install the debugger from here
2. Open the debugger in VSCode.
3. Set break points in the source code (*.ts).
4. Clicking on the "Start Debugging" button will launch the Chrome browser using the configuration in launch.json.