Welcome to the transformative world of Natural Language Processing (NLP)! In this article, we’ll dive into the details of the Question Generator system, a powerful tool designed to generate reading comprehension-style questions from a variety of text sources, like news articles and book excerpts. Let’s explore how to set it up and use it effectively!

Getting Started: Installation and Setup

To begin your journey with the Question Generator, you’ll need to set up the environment. Here’s a step-by-step guide:

1. Clone the repository:

git clone https://github.com/amontgomerie/question_generator

2. Navigate to the project directory:

cd question_generator

3. Install the required packages:

pip install -r requirements.txt -qq

Generating Questions: The Basic Usage

The easiest way to generate questions is by running the run_qg.py script. Specify the text file that contains the content you wish to generate questions from:

python run_qg.py --text_file articlestwitter_hack.txt

This command will generate 10 question-answer pairs based on the text in the specified file and print them to your console.

Question Generation through the QuestionGenerator Class

Alternatively, if you’re interested in a more programmatic approach, you can leverage the QuestionGenerator class directly as follows:

from questiongenerator import QuestionGenerator
qg = QuestionGenerator()
questions = qg.generate(text, num_questions=10)

In this case, you simply plug in the text from which you want to generate questions, and the function will return a list of question-answer pairs!

Understanding the System: An Analogy

Think of the Question Generator as a chef preparing a meal. The text you provide is akin to raw ingredients. The Question Generator model is like a skilled chef who knows how to transform those ingredients into delicious dishes (questions and answers). The QA Evaluator acts as a food critic, sampling the dishes and deciding which ones are the best based on taste and presentation (the quality of question-answer pairs). If you have too many ingredients (questions), the chef may struggle, compromising the quality of the final dishes! Therefore, it’s about balancing quality and quantity.

Choosing the Right Number of Questions

You have control over how many questions the model generates, which you can adjust using the --num_questions command-line argument or directly in the generate method:

qg.generate(text, num_questions=20)

Be mindful that if you request too many questions, the output quality may diminish. The model is tuned to deliver the best results based on the length of your input text!

Answer Styles: Tailor Your Output

The Question Generator can craft questions with various answer styles:

• Full-sentence answers
• Multiple-choice answers
• A mix of both styles

--answer_style mixed