情報処理学会 第77回全国大会講演要旨

After acquired dish name recognition results, we propose a knowledge base (KB) based pipeline method with semantic inference to realize Dish-Ingredient relation extraction. Here, Knowledge base is used to infer Dish-ingredient relation path and text feature selection method is to compute ingredient score. Only the ingredients which are more than threshold can be reserved in the paths. Finally, we acquire accurate D-I relations by traversing and computing feature scores.