Telegram Chat Bot Using Machine Learning

Description:

This is "just for fun" project based on SKlearn ML models.
The main idea is try to create a Russian language talkative bot which determines user's main intention based on the input phrase and give the most appropriate answer to him.
A ready-made dictionary of intentions and examples of answers to them was used for bot training (JSON).

Models:

Three different models were tested to determine the user's intention in his phrase. Results are below:

• Logistic Regression (model score = 0.3884)
• Random forest Classifier (model score = 0.8281)
• MLP Classifier (model score = 0.8247)
  So, I decided to use Random forest Classifier because it is faster and less GPU expensive.

Input Data Preparation filter_text():

The input word (phrase) is lowered to lower case then spaces and punctuation marks are removed from it.

Word Comparison is_match():

Two words are compared: the input and which the model predicts in the body of this function.

Main Logic Func bot():

1. Filter input data;
2. Try to find the answer directly in the dictionary;
3. If not - use ML model intent predictiction and take the random answer example from this intent group;
4. Or use the Failure Phrases if the model score is not enough in any case.
5. Repit until input phrase == one of exit_phrases.

Bot conversation example:

Server output:

As you can see, users names are shown.

Telegram Output: