This repository contains materials for the AIST conference.

Both datasets have been manually anonymized due to VK's privacy policy. Besides, they were normalized in a bit different way comparing with their initial versions, used in this study.

msg_with_emoji.csv is a small dataset (~10k samples), containing annotated messages with emoji.

train_data.csv is a relatively large dataset (~54k samples; only half of the dataset is uploaded here, full version can be uploaded later).

model.joblib is a trained logistic regression model for predicting emotions in Russian text messages.

vect.joblib is a TF-IDF vectorizer for the aforementioned model.

Parser.py is a high-level parser for text messages requested from VK and Telegram.

It requires the path to the folder with messages and recursively iterates through all files.

VK parser looks into *.html files and TG parser checks *.json files.

from Parser import Parser
parser = Parser('PATH', mtype='tg') # parse TG messages
parser = Parser('PATH', mtype='vk') # parse VK messages
parser.parse() # returns a DataFrame with text messages