It's a very basic Chatbot for Python Django including NLTK and Django-REST-framework. This Chatbot is currently working without Machine learning algorithms. Decisions are made by simple statistic evaluation.
The Algorithm is based on labeled data on your Django Database and the tool is supporting continuous labeling.
- Python (3.7, 3.8, 3.9)
- Django (2.2, 3.0, 3.1, 3.2)
Install using pip
...
pip install django-simple-chatbot
add simple_chatbot
to your INSTALLED_APPS
setting.
INSTALLED_APPS = [
...,
'simple_chatbot'
]
Note: Make sure to run manage.py migrate
after changing your settings.
The simple_chatbot app provides Django database migrations.
Create a response.py
file inside of an already installed app.
from simple_chatbot.responses import GenericRandomResponse
class GreetingResponse(GenericRandomResponse):
choices = ("Hey, how can I help you?",
"Hey friend. How are you? How can I help you?")
class GoodbyeResponse(GenericRandomResponse):
choices = ("See you later.",
"Thanks for visiting.",
"See ya! Have a nice day.")
Add this Response to your SIMPLE_CHATBOT
setting
SIMPLE_CHATBOT = {
...
'responses': (
("YOUR_APP.responses.GreetingResponse", "Greeting"),
("YOUR_APP.responses.GoodbyeResponse", "Goodbye"),
),
}
Go to your Django admin and create greeting
and goodbye
tags.
Your response options will be selectable via choices.
Go to your Django admin, write some patterns and label them. You can just use the following labels:
[Greeting]
"Hi, how are you?", "Is anyone there?", "Hello", "What's up?!", "hey there!"
["Goodbye"]
"Bye", "See you later", "Goodbye", "I need to go now."
Note If you do not want to write that patterns by yourself, use a command
manage.py simple_chatbot_initial
. You need to label them after initializing.
The package will automatically tokenize the input and map tokens to labels.
Add simple_chatbot url to your routings:
from simple_chatbot.views import SimpleChatbot
urlpatterns = [
...
path("simple_chatbot/", SimpleChatbot.as_view())
]
Make a Post request to your new endpoint:
curl \
-H "Content-Type: application/json" \
--data '{"message":"how r u?"}' \
http://localhost:8000/simple_chatbot/
The response should look like
{
"tag": "Greeting",
"message": "Hey, how can I help you?"
}
Pattern
- message which might be send by a user. Add a tag to the pattern for being able to identify and response to that messageTag
- includes information about Response class for a specific methodToken
- tokenized words which are referencing to different patterns. The user-input will be identified by different tokens.UserMessageInput
- new inputs from production. It contains information about chosen pattern. You can label that messages later and include them into the system.
You can add following options to your SIMPLE_CHATBOT
setting:
- STEMMER_MODULE: nltk package for stumming your strings -
default:
nltk.stem.lancaster.LancasterStemmer
. - responses: choices for your tags. It should reference to a response class. Warning You won't be able to create tags without response classes.
The simple_chatbot.responses
package provides currently following response classes:
- BaseResponse
- GenericRandomResponse
It's just an abstract class for require a specific shape of your response classes. If you are creating a new response, you should inheritance from that class.
It will choose a generic answer from class property choices
.
This simple_chatbot.responses
includes a single view SimpleChatbot
.
This view is reusable. The most important changeble option:
save_pattern
: if True each message will be saved and you can post label the incoming messages. defaultTrue
.
- Required Request type:
POST
- payload:
{message: "YOUR MESSAGE"}
- response:
{tag: 'TAG', message: 'RESPONSE'}
It's a very basic Chatbot decisions are made with tools by NLTK which follows basic preprocessing for NLP of tokenization and stumming.
This package is inspired by the Chatbot Tutorial of Tech witch Tim. Checkout his blog: https://www.techwithtim.net/tutorials/ai-chatbot/
In my opinion the used script is teaching important concepts but it's overtooled by using Deep-Learning algorithm on that small amount of data.
Real NLP's and Deep Learning algorithms needs a large amount of data. One problem in smaller and beginning projects: You won't have that amount of data by starting your projects.
This package gives you possibilities to work with a small amount of data and it helps you to collect new data for being able to use deep learning algorithms one day.
Fork the repo and get stuck in!