
This will make it convenient to include these bots in Zulip API releases on pypi. Fix #5009.
31 lines
1.3 KiB
Markdown
31 lines
1.3 KiB
Markdown
John
|
|
|
|
Instructions:
|
|
You'll have to install chatterbot to use this bot.
|
|
Please run: pip install chatterbot on your command line.
|
|
The script will need to download some NLTK packages after running in your
|
|
home directory. With the mission of humanizing bot interactions, John aims to be your
|
|
virtual assistant at the hour of asking for help in Zulip. John is an
|
|
interactive bot that uses machine learning heuristics to simulate a
|
|
conversation with the user. He has a great sense of humor and
|
|
is also powered by Open Source code!
|
|
|
|

|
|
|
|
How it works?
|
|
John is initially trained with Corpus files, or large text files.
|
|
Dialogues are loaded into a json "database", he will try to follow them
|
|
once it receives input from a user. John will query the database and
|
|
try to find the response that best matches the input according to the Levenshtein distance
|
|
which is a string metric for measuring the difference between two sequences. If several
|
|
responses have the same acurracy, he will choose one at random.
|
|
|
|

|
|
|
|
Can he learn by himself?
|
|
John's engine allows him to learn from his conversations with people. However,
|
|
without strict supervision bots that learn from people can do harm, so learning
|
|
is currently restricted to his initial corpus.
|
|
|
|

|