Sequence classification model for sentiment analysis of Russian conversational text

Project structure

notebooks

.ipynb files

• csv_preprocessing.ipynb - pre processing initial csv files
• eda.ipynb - exploratory data analysis
• training.ipynb - model training
• evaluation.ipynb - model evaluation

src

.py files

• training.py contains Trainer class to train BertForSequenceClassification
• prediction.py contains Predictor class to predict sentiments from text data using pre trained model
• metrics.py contains functions to compute precision, recall, and f1 scores
• evaluation.py contains a function to evaluate pre trained model using test data
• input_data_preprocessing.py contains utility function to fix initial csv files

Metrics

	Positive	Neutral	Negative	Average
Precision	0.744	0.717	0.799	0.753
Recall	0.811	0.623	0.833	0.756
F1	0.776	0.667	0.816	0.753