RecipeMate: A Food Media Recommendation System Based on Regional Raw Ingredients
A Padmavathi, Dipta Sarker
Abstract
This research paper explores a recipe recommendation system that suggests recipes based on input ingredients, cuisine, and undesired ingredients. RecipiMate uses natural language processing alongside other machine learning techniques to pre-process and analyze a dataset with about 28K recipes. The dataset was cleaned and pre-processed by removing missing data, converting the ingredient lines to a list of ingredients, and lemmatizing the ingredients. The resulting dataset was then analyzed using cosine similarity to calculate pairwise similarities between all recipes. The resulting similarity matrix was then used to recommend the best recipes based on the input ingredients, cuisine, and excluding any unwanted ingredients which might be allergic to a person. The system was developed using Python and several libraries, including Flask, Pandas, NumPy, NLTK, and Scikit-learn. The resulting system was deployed on a Flask web application and manually tested. The results show that the system provides accurate recipe recommendations, improving the user experience of recipe search and discovery.