Skip to content

JersonGB22/QuestionAnswering-TensorFlow

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Question Answering

This repository implements Question Answering models, a natural language processing (NLP) task that retrieves the answer to a question from a given text. These models are created using the TensorFlow and Hugging Face Transformers libraries. There are two types of this task:

  1. Extractive: Extracts the answer directly from the provided context.
  2. Abstractive: Generates a correct answer freely based on the context or without it.

Common Use Cases:

  • Virtual Assistants: Answering user questions using information from a specific database.
  • Document Search: Finding precise answers in large document sets.
  • Customer Service: Providing quick and accurate responses to customer inquiries.
  • Education: Helping students get answers to questions based on study materials.

Implemented Models:

  • Extractive Model for Answerable Questions: This RoBERTa Large model has been fine-tuned with the Stanford Question Answering Dataset (SQuAD) to appropriately answer questions within a given context. SQuAD consists of 100,000 question-answer pairs across more than 500 Wikipedia articles, where the answer is a text segment from the corresponding passage. The model has achieved an excellent F1 Score of 94.1% on the validation set.

  • Extractive Model for Answerable and Unanswerable Questions: This model is a more robust and challenging version of the previous one, fine-tuned with SQuAD 2.0 to appropriately answer both answerable and unanswerable questions within the provided context. SQuAD 2.0 consists of the 100,000 question-answer pairs from SQuAD plus 50,000 unanswerable questions. The model is capable of responding to questions within the context and also returns No Answer if the answer is not in the context or if it is of an abstractive type. It has achieved an excellent F1 Score of 87.7% on the validation set.

Some Results




Further results can be found in their respective notebooks.

Technological Stack

Python TensorFlow Hugging Face Plotly

Contact

Gmail LinkedIn GitHub