Is it possible to classify a recipe’s cuisine type just from a list of ingredients? Our project seeks to answer this question and does so using some basic tools of Natural Language Processing. We take a dataset from Kaggle.com that has a list of ~40,000 recipes with a cuisine type classification. One method we employ is called a Bag of Words (BoW) model which take all words found in the ingredients list and builds a classifier based on the occurrences of those words in the training set. The other method is the Term Frequency-Inverse Document Frequency (TF-IDF) which considers the frequencies of individual words in the training set. Both methods produced a testing accuracy of greater than 60%, which is good considering that we implemented the most naïve NLP models.
The included Jupyter Notebooks showcase our approaches to solving this problem. Another person from Kaggle solved this problem with a 78.8% accuracy using the NLTK toolkit's built-in TF-IDF vectorizer (https://www.kaggle.com/code/rahulsridhar2811/cuisine-classification-with-accuracy-78-88/notebook). We attempted some simpler models that are more easily understood for the purposes of learning the basics of NLP, so we speculate that a more advanced classifier could be easily capable of producing more than 80% accuracy.