Boolean Question Generation and Answering System for Cooking Recipes
Combine boolean question generation and boolean question answering, to answer the generated questions. The dataset will be annotated manually.
Our datasets and models codebases: the dataset used is Food Ingredients and Recipes Dataset with Images : https://www.kaggle.com/datasets/pes12017000148/food-ingredients-and-recipe-dataset-with-images the models codebases are
Our evaluation base model : https://pypi.org/project/boolean-question/
roberta-base : https://huggingface.co/roberta-base
roberta-base-boolq : https://huggingface.co/shahrukhx01/roberta-base-boolq
To run the code download the models and place them in models folder and the datasets exist in our repository then run the desired notebook
trained and stored model:
1.roberta-base retrained with 2000 auto-generated questions : https://drive.google.com/file/d/1oRte7R7lthNWoUWnfSB-JHP7sqcyoKsE/view?usp=sharing
2.roberta-base-boolq retrained with 2000 auto-generated questions : https://drive.google.com/file/d/1vnJQAz2DsTiRCqG68I3ygjwmecvxt5Ep/view?usp=sharing
Our questions datasets are avalible in our git repository
Based on our intial evaluations, base model's accuracy is 55.5% when tested with auto-generated questions and 64% when tested with manually-generated questions, and the first model's accuracy is 55.5% when tested with auto-generated questions and 57.1% when tested with manually-generated questions, and the second's accuracy is 61.75% when tested with auto-generated questions and 61% when tested with manually-generated questions.
Most of our datasets collecting/building/pre-processeding are illustrated in our QuestionGeneration file. Our data spilting is demonstrated in Model_retraining file.
Our solution architecture, how you provided parameters and how you supplied it with training data are demonstrated in Model_retraining file.
Our evaluation is represented in EvaluateBaseModel file for the base model, and EvaluateOurModels file for our models.
Mainly we will be using Demo file to Demonstrate our work.