rabidgremlin/Concord

Web based tool for crowd sourcing labeling of training data for NLP. (WIP)

Concord

concord. noun. agreement or harmony between people

Concord is a web application designed to easily crowd source labelling of text data for NLP. It's focus is on labeling text for training text classification models such as in chatbots.

get up and running (for developers)

1. Create database

docker run --rm -it -p 3306:3306 -e MYSQL_DATABASE=concorddb -e MYSQL_USER=concorduser -e MYSQL_PASSWORD=concordpwd -e MYSQL_ROOT_PASSWORD=password123 -d mysql:5.7

2. Setup RASA NLU Server

By default Concord is configured to use a Rasa NLU server for testing. See concord-server/server/src/main/yml/server.yml

2a. Start server

docker run -p 5000:5000 rasa/rasa_nlu:latest-full

2b. Populate RASA NLU server

In a new terminal window:

curl -XPOST --header "Content-Type: application/json" http://localhost:5000/train?project=taxibot -d @testbed/taxibotdata.json

2c. Check status of model build

curl http://localhost:5000/status

When status of taxibot model is listed as ready, execute:

curl -XPOST http://localhost:5000/parse -d '{"q":"get me my taxi", "project": "taxibot"}'

NOTE: Wait for this response to return before running other queries against the Rasa NLU server. May take several minutes.

3. Create database and Start up server

3a. Create the database

cd concord-server/server
./gradlew createDatabase

3b. Run server

cd concord-server/server
./gradlew run

4. Load test data

In a new terminal window:

4a. Get Token for API calls

By default Concord is configured to use a list of usernames/passwords held in it's config file. See concord-server/server/src/main/yml/server.yml

curl -v -X POST http://127.0.0.1:9000/api/sessions --header "Content-Type: application/json" --data '{"userId":"Bob","password":"secret"}'

Extract token from response and use in place of XXXXXX in the following commands.

4b. Load classification labels

curl -v -X POST http://127.0.0.1:9000/api/labels/bulk --header "Content-Type: text/csv" --header "Authorization: Bearer XXXXXX" --data-binary '@testbed/labels.csv'

4c. Load unlabelled phrases

curl -v -X POST http://127.0.0.1:9000/api/phrases/bulk --header "Content-Type: text/csv" --header "Authorization: Bearer XXXXXX" --data-binary '@testbed/unlabelled_phrases.csv'

5. Install and start up front end

In a new terminal window:

5a. Install dependencies

cd concord-ui/
npm install

5b. Start up front end

cd concord-ui/
npm start

5c. Login

Login as bob, password secret or as alice, password garden

---

Viewing your Database

docker exec -it [docker-image-name] bash

mysql -u concorduser -p

password: concordpwd

use concorddb