Extraction of ingredient names from recipes by combining linguistic annotations and CRF selection

Nutrition and diet have direct and considerable impact on our well-being and health. This field attracts researchers from different areas, such as medicine, nutrition and epidemiology, computer sciences, artificial intelligence and natural language processing (NLP).We process the recipes with NLP methods in order to automatically identify ingredient names within recipes. We propose a hybrid system based on linguistic enrichment of the recipes and selection of the relevant ingredient names with a CRF method. Semantic resources have been specifically built for processing two kinds of information: exact (e.g. quantity expressed in grams or liters, durations expressed in minutes or days) and fuzzy (e.g. quantities expressed in chouilla (smidgeon) and louche (ladle), durations sequenced with aprèes, ensuite, alors que (the, after that, while)). The experiments are performed with French-language textual data. The results demonstrate that the proposed method is useful for searching and managing the recipes.