pip install spacy
python -m spacy download en_core_web_lg
pip install https://github.com/Robstei/chatbotsclient/releases/download/1.0.0/chatbotsclient-1.0.0.tar.gz
pip install -U https://github.com/Robstei/chatbotsclient/releases/download/1.0.0/chatbotsclient-1.0.0.tar.gz
This package consists of a Moderator and a Chatbot class to make chatbots talk to each other. It is required to have a moderator instance up running before chatbots try to connect to the conversation. Messages are sent through websocket channels using pusher. The moderator collects all messages from connected chatbots and selects the best fit.
Simply instantiate a Moderator object. The moderator will wait for chatbots to connect and provides the possiblity to start the conversation by input.
The chatting-chatbots repository already consists of a moderator script: chatting-chatbots/moderator/moderator.py.
# chatting-chatbots/moderator/moderator.py
from chatbotsclient.moderator import Moderator
moderator = Moderator()Before connecting your chatbot to the conversation wait for the moderator to prompt "Message: ". Otherwise the connection might not be established successfully. Once chatbots are connected, the conversation may be initialized by inputing a string.
While the conversation is ongoing the moderator script will prompt message scores. Chatbots will only respond to messages of other chatbots.
Instantiate a Chatbot object and pass your custom respond function. When ever a message from the moderator is received the provided respond method will be executed. The moderator script must run in first place.
from chatbotsclient.chatbot import Chatbot
from chatbotsclient.message import Message
from typing import List
def compute(message, conversation):
# custom answer computation of your chatbot
...
def respond(message: Message, conversation: List[Message]):
answer = compute(message.message, conversation)
return answer
chatbot = Chatbot(respond, "<chatbot_name>")The compute method is meant as a placeholder for your specific method to return an answer for the provided message. Thus the method is not part of chatbotsclient package.
You may also ignore the conversation list:
def compute(message, conversation):
# custom answer computation of your chatbot ignoring the conversation list
...
def respond(message: Message, conversation: List[Message]):
answer = compute(message.message)
return answer
A Message object is passed to the custom respond function of your bot. It contains the plain text message as well as information about the sending bot.
| Field | Description |
|---|---|
| message | Plain text message. Used to compute your chatbots answer. |
| bot_id | Id of the sending bot. |
| bot_name | Name of the sending bot. Could be used for entity replacement. |
List containing all previously selected messages.
The moderator ranks all incoming message by following criteria:
| Method | Description |
|---|---|
| Similarity | To fit the previous message, but also to prevent looping conversations |
| Conversation share | To ensure a varied conversation |
| Topic | Subject fitting |