Food Classification for Inflammation Recognition Through Ingredient Label Analysis: A Real NLP Case Study

Abstract

As of late, literature shows that food intolerances affect a large portion of the world population. Diagnosis and Prevention are essential to avoid possible adverse responses due to food ingestion. Concerning this point, consumers and industry players are also demanding tools useful to warn individuals about the composition of commercial products. In this scenario, Natural Language Processing (NLP) approaches can be very useful to classify foods into the right intolerance group given their ingredients. In this work, we evaluate and compare different deep and shallow learning techniques, such as Linear Support Vector Machine (Linear SVM), Random Forest, Dense Neural Networks (Dense NN), Convolutional Neural Networks (CNN), and Long short-term memory (LSTM) with different feature extraction techniques like Bag of Word (BoW), Term Frequency-Inverse Document Frequency (TF-IDF), and Word2Vec, in order to solve this task on real commercial products, aiming to create a baseline for future works and a software-product. In the end, interesting and noticeable results have been achieved and the baselines have been identified into the Linear SVM and the Dense NN with Bag of Words or with the combination of Bag of Words, TF-IDF and Word2Vec.