Skip to content

A deep learning model for classification of Judgemental and Non-Judgemental statements

License

Notifications You must be signed in to change notification settings

hardiksraja/Textual-Judgemental-Analysis

Repository files navigation

Textual-Judgemental-Analysis

A deep learning model for classification of judgemental and non-judgemental statements; to overcome challenges of identifying factual content, compared to fictional/nonfactual content.

We have solved this problem using a pre trained model called DistilBERT. The output of DistilBERT model is used in Logistic Regression Model to predict the class. DistilBERT is a transformer based model which is used for classification of text.

We have used two models here :
Model Selection : Supervised Learning > Binary Class Classification > DistilBERT > Logistic Regression
DistilBERT processes the sentence and passes along some information it extracted from it on to the next model. DistilBERT is a smaller version of BERT developed and open sourced by the team at HuggingFace. It’s a lighter and faster version of BERT that roughly matches its performance.

The next model, a basic Logistic Regression model from scikit learn will take in the result of DistilBERT’s processing, and classify the sentence as either Judgemental or Non Judgemental (0 or 1, respectively).

The data we pass between the two models is a vector of size 768. We can think of this of vector as an embedding for the sentence that we can use for classification.

Refer, Judgemental_Non_Judgemental_Anjali_Hardik_Jayant.pptx for detailed description on the project

Model Evaluation Metrics – Accuracy and ROC AUC Score

We built 4 different models and finally selected DistilBERT

ModelComparison

Deployment/Demo

Deployed at : https://judgement-prediction-streamlit.herokuapp.com/

ModelDeployment

Deployment Files: The models were deployed on Heroku with UI built via StreamLit

Refer, 2021-07-14_StreamLit & Heroku_HardikRaja.pptx for commands and background on Deployment


judgement_predictor.pkl - Saved Logistic Regression pickle file Procfile - Executing the StreamLit App requirements.txt - Required by Heroku setup.sh - File required to deploy StreamLit App on Heroku streamlit_app.py - The StreamLit App

About

A deep learning model for classification of Judgemental and Non-Judgemental statements

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published