Information Extraction from Unstructured Recipe Data

Abstract

Online food recipes are an important source of information for many individuals, who use these to learn how to cook new dishes and choose their meals. However, these often lack structured information, useful to improve search and recommendation systems of food recipe websites, as well as calculate accurate nutritional information, which brings additional value to users. To solve this problem, FRIES was developed. FRIES automatically extracts the names, quantities, units and cooking methods for each ingredient in a recipe. The system uses mainly rule-based methods and achieves an average F-measure of 0.89 for the extraction of the cooking methods present in a recipe and an average F-measure of 0.83 for the extraction of associations linking cooking methods to ingredients. FRIES' results show that it can accurately and automatically extract information from cooking recipes. This information can be used to estimate the nutritional information of food recipes and support recommendation systems.

Towards Multi-Language Recipe Personalisation and Recommendation

“…Extracting the quality and quantity of recipes and ingredients [29,32] is a key precursor in many application areas of food computing, including healthy recommendation [33]. The multi-modal aspect of recipes has shown promise in enhancing cooking procedure understanding [40] by using auxiliary data such as video [22,30] or images [27,42].…”

Hidden Markov Based Mathematical Model dedicated to Extract Ingredients from Recipe Text

“…Extracting ingredients automatically from a recipe text is an extremely useful activity especially when we want to analyze a massive data of text recipes. Rule-Based methods were implemented to extract information from unstructured recipe data (Silva, Ribeiro, & Ferreira, 2019) Ingredients is not the only useful information we want to extract; in this work we are going to use Hidden Markov Models especially Viterbi algorithm with some modification to make it receiving two unique features: POS-tags and tokens, to predict ingredient states.…”