Publication

Abstract : In today’s world where an individual is becoming more and more busy and independent, the use of recommendation-based systems is steadily increasing. Thus, making available professional knowledge to the common man in a short-span quite necessary. The aim of our recipe recommendation system is to recommend recipes to users based on their questions. To make the recommendation model important as well as meaningful, it is pertinent to display only those recommendations that have a greater probability to be fit for the asked question. We have addressed this challenge by working on a threshold parameter generated from the recommendation engine. Apart from this, we have also included a question classification (QC) task together with the question answering (QA) module. The QA module is used to extract the requisite answers from the recommended recipe based on the class label obtained from QC. The main contribution of this work is the proposal of a robust recommendation approach by enabling analysis of threshold estimation and proposal of a suitable dataset. The final output of the recommendation system obtains benchmark results on the human evaluation (HE) metric. Our code, pretrained models and the dataset will be made publicly available.