A proof of concept autonomous conversational agent for my honours project.
To run this, your system has to satisfy the following requirements:
- Python >=2.7.2, but not 3
- Java Hotspot VM (Java 5 or above)
- BeautifulSoup 3.2.1
- foreman
- at least 512MB RAM to load the classifiers into memory
- Clone the project and
cd
into the project directory. mv .env.example .env
and change the file with your settings.- Edit
identity.py
to add identities to the agent. foreman start
/classifiers
contains jar binaries for the Stanford MaxEnt and CRF classifiers./data
contains a serialization of the current state of the agent/incoming
emails that will/have been processed/misc
various notes on the classifiers which I don't want to forget/models
contain models for the MaxEnt and CRF classifiers/scenarios
the knowledge base of the agent/scripts
various quick-and-dirty scripts that I have had to write at some point. Not recommended for reuse..
contains the main components
- Crawlers and parsers to automatically obtain new messages and responses
- Multiple agent identities
- Email threading
- Classifiers to recognize message type, main actors, etc.
- Sophisticated AI for response generation
- Metrics module