ChatterBot backend

[yuvalBor]

1. Can you please shed some light on how chatter bot training works and how it chooses the utterance for a response.

I mean, do you use Neural networks, ML, deep learning or just some rule based (string distance -
levenshtein distance) methodologies?

1. Is it contextual?

Thanks,
Yuval

[sal48]

So I just started delving into the backend, and I can share what I know with you. If you wanted to know the exact lines of code, you can find the relevant code in chatterbot.generate_response method. Basically, when you setup your chatbot instance, you should setup the logic adapters you want to use:

bot = ChatBot('Ava',
    logic_adapters=[
        {
            "import_path": "chatterbot.logic.BestMatch",
        "statement_comparison_function": "chatterbot.comparisons.levenshtein_distance",
            "response_selection_method": "chatterbot.response_selection.get_first_response"
        }
    ],
    filters=[
        'chatterbot.filters.RepetitiveResponseFilter'
    ]
)

In the above snippet, I use Levenshtein Distance for statement comparison, but I also use the Best Match algorithm for lookups, as well a last adapter for response selection. These logic adapters will be used after the preprocessor code (trim white space, escape chars, anything else you added) is executed.

Hope this helps!

[vkosuri]

1. Neural networks :heavy_multiplication_x:
2. ML :heavy_check_mark:
3. Deep learning :heavy_multiplication_x:
4. just some rule based (string distance -levenshtein distance) methodologies :heavy_check_mark:
5. contextual :heavy_check_mark:

[lock[bot]]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.