PREDICT CLOSED QUESTIONS ON STACK OVERFLOW

Project dependencies

This project aims to develop a machine learning model using Bidirectional Encoder Representations from Transformers (BERT) to predict whether a question posted on Stack Overflow will be closed. This can assist moderators in identifying potentially low-quality questions and improve the overall platform's efficiency.

Requirements

• Python 3.6+
• Necessary libraries (install using pip install <library_name>)
  • transformer
  • torch (GPU-accelerated training highly recommended)
  • pandas
  • numpy
  • sklearn (for preprocessing)

DATA

• The code assumes you have access to a Stack Overflow dataset containing questions (textual content), labels indicating whether they were open, not a real question, not constructive, too localized or off topic, and potentially additional features like tags, timestamps, or user information.
• Download or prepare your dataset accordingly.

Model Architecture

• This code utilizes a fine-tuned BERT model for text classification. BERT effectively captures contextual relationships within text, making it well-suited for this task.

This is the architecture on which the model is trained

Code Structure

• tokenizer.py: This Python file (tokenizer.py) is designed to implement a tokenizer function that breaks down text strings into smaller units called tokens. These tokens can be words, punctuation marks, characters, or n-grams (sequences of n words) depending on the specific tokenization method employed.
• preprocessing.py: This is a Python file (preprocessing.py) which is designed to implement on to extract all the important columns in the given dataset and to concat to form a single column containing all the important information and the tokenization is applied on this column and this is send as input in input layer of BERT