Recently, the number of user-generated recipes on the Internet has increased. In such recipes, users are generally supposed to write a title, an ingredient list, and steps to create a dish. However, some items in an ingredient list in a user-generated recipe are not actually edible ingredients. For example, headings, comments, and kitchenware sometimes appear in an ingredient list because users can freely write the list in their recipes. Such noise makes it difficult for computers to use recipes for a variety of tasks, such as calorie estimation. To address this issue, we propose a non-ingredient detection method inspired by a neural sequence tagging model. In our experiment, we annotated 6,675 ingredients in 600 user-generated recipes and showed that our proposed method achieved a 93.3 F1 score.

This study presents a voice assistant system to support cooking by utilizing smart speakers in Japan. This system not only speaks the procedures written in recipes point by point but also answers the common questions from users for the specified recipes. The system applies machine comprehension techniques to millions of recipes for answering the common questions in cooking such as “人参はどうしたらよいですか (How should I cook carrots?)”. Furthermore, numerous machine-learning techniques are applied to generate better responses to users.

In a recipe sharing service, users publish recipe instructions in the form of a series of steps. However, some of the “steps” are not actually part of the cooking process. Specifically, advertisements of recipes themselves (e.g., “introduced on TV”) and comments (e.g., “Thanks for many messages”) may often be included in the step section of the recipe, like the recipe author’s communication tool. However, such fake steps can cause problems when using recipe search indexing or when being spoken by devices such as smart speakers. As presented in this talk, we have constructed a discriminator that distinguishes between such a fake step and the step actually used for cooking. This project includes, but is not limited to, the creation of annotation data by classifying and analyzing recipe steps and the construction of identification models. Our models use only text information to identify the step. In our test, machine learning models achieved higher accuracy than rule-based methods that use manually chosen clue words.