This project provides a convenient environment and automation pipeline to
- Construct trainable conversation data from Facebook Messenger chat history
- Train chatbot models of individual friends based on conversation data
- Wrap these models for simple deployment to fit third-party chatbot API
- ready - config demonstrated via DeepQA-trainable execution
- ready - successful batch training demonstrated via DeepQA module
- ready - automated inferrence of trainable sentence lengths via messenger statistics
- todo - need to script deployment and prepare example wrapper
Assuming that all external dependencies are ready and respective paths appropriately specified under the config subdirectory, the training and deployment execution scripts can be run immediately from the project root.
Of course, the environment setup has some fairly strict requirements:
- Ensure that your local environment is equipped with a Bash version >=3, in addition to
seperate distributions of Python 2.7+ and Python 3+ (ideally managed by conda)
- Install the fb-chat-archive-parser via pip under the Python 2.7 environment by running:
pip install fbchat-archive-parser- Install some training library to which the friendbot factory will relay conversation data for consumption
(the default is DeepQA)
- Ensure that an uncompressed
facebook archive
is made available
(conventionally under data/facebook_unstructured)
- Specify desired setup at config/training/training.config & config/deployment/deployment.config
- Run
app/bash/train.sh,app/bash/stats.sh, orapp/bash/deploy.shfrom project root per relevant task
This code is under the Apache License 2.0.
If you use or modify fb-friendbot-factory, please credit the original author as
- Logan Martel - https://github.com/martelogan