This project is a bot for Telegram, which can find answers to questions based on the information from provided sources. The bot uses an artificial neural network and NLP techniques to find the most relevant answer.
- Python 3.6+
- pip
- virtualenv
- git
- Telegram bot token
- model file (you could use the provided one)
- navec_news_v1_1B_250K_300d_100q.tar
- ru_small.bin (download from here and unpack)
- Clone the repository
git clone <repo_url>- Create a virtual environment and activate it
virtualenv venv
source venv/bin/activate- Install the requirements
pip install -r requirements.txt- Create a file named
.envin the root directory of the project and fill it with the following data:
API_TOKEN=your_telegram_bot_tokenObtain the token from @BotFather
- Run the bot
python bot.pyThe bot is able to answer questions about the following topics:
- SSTU (information for abiturients to help them get into the university, information about the university itself)
The bot uses a neural network to find the most relevant answer to the question. The model is trained on the dataset.json dataset. The model is trained using the Dmitry Korobchenko algorithm. You can train the model yourself using the training.ipynb notebook. Don't forget to provide your own dataset with most releted questions in your case.
The bot uses Navec and Natasha it self to get the vector representation of each word in the question and them sum them up to get the vector representation of the whole question. You can download the Navec model from here or train it yourself in order to increase the accuracy (see this for more information).
The bot uses JamSpell to correct the spelling of the question. However, the model is trained on a piece of Russian text, so it may not work well with some words. You can download the model from here or train it yourself (refer to this).
The bot is able to answer questions only in Russian because of the NLP models and tokenizer used in Natasha.