https://github.com/adamspannbauer/app_rasa_chat_bot
a stateless chat bot to perform natural language queries against the App Store top charts
https://github.com/adamspannbauer/app_rasa_chat_bot
chatbot dash nlp nlu plotly rasa
Last synced: 2 months ago
JSON representation
a stateless chat bot to perform natural language queries against the App Store top charts
- Host: GitHub
- URL: https://github.com/adamspannbauer/app_rasa_chat_bot
- Owner: AdamSpannbauer
- License: mit
- Created: 2018-03-05T19:33:49.000Z (over 7 years ago)
- Default Branch: master
- Last Pushed: 2018-03-28T16:24:49.000Z (over 7 years ago)
- Last Synced: 2025-04-08T16:38:05.691Z (6 months ago)
- Topics: chatbot, dash, nlp, nlu, plotly, rasa
- Language: Python
- Homepage:
- Size: 22.6 MB
- Stars: 29
- Watchers: 1
- Forks: 13
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
## Rasa NLU to query App Store Top Charts
[Related blog post](https://adamspannbauer.github.io/2018/03/15/building-a-rasa-chatbot/) with more in depth write up of the process used in this repo.
### Output
The above image is a screen cap from a limited [Plotly Dash](https://plot.ly/products/dash/) app created to interact with the bot. The app code is in [`dash_demo_app.py`](dash_demo_app.py).
The file [`live_test_rasa.py`](live_test_rasa.py), allows a similar experience from the command line instead of via Dash in browser.
### Input Data:
* [`data/generic_rasa_train_data.json`](data/generic_rasa_train_data.json): taken from the rasa intro restaurant chatbot example; all of the restaurant intent examples were removed
* [`data/app_chart_data.csv`](data/app_chart_data.csv): table of top chart apps; the relevant column is just the list of app names to use as entities in training (table was created by [`utils/downloader.py`](utils/downloader.py))
### Training process:
* Generated domain specific training data with [`gen_training_data.py`](gen_training_data.py) and [`generic_rasa_train_data.json`](data/generic_rasa_train_data.json)
* Parameterized phrases were created to fill in the blanks with randomly chosen entities
* eg: `'show me the {ordrank} most popular {chart} app'`
* Created `N` variations of the parameterized phrases and added them to the generic training data
* output saved to [`data/app_train_data.json`](data/app_train_data.json)
* Train the rasa model
* [`train_rasa.py`](train_rasa.py) (generic train script from rasa docs)