👨‍👩‍👦

food2vec – Augmented cooking with machine intelligence

Entry
food2vec – Augmented cooking with machine intelligence
Simple Title
food2vec – Augmented cooking with machine intelligence
Type
demo
Year
2017
Posted at
January 27, 2017
Tags
essential
image

Overview

名前の通り、word2vecの考え方を、料理の材料に当てはめたもの.

Abstract

Machine intelligence has changed your life, from how you listen to music through Discover Weekly playlists, consume news through Facebook, or talk to your hand computer’s friendly digital assistant. But why hasn’t it changed how we eat? Can we modify the ingredients of language processing algorithms to get insights about food? If you tell me what you want to eat, can I recommend complementary foods, much like Spotify recommends complementary songs?

Technology/System

下記のインタラクティブなデモは95896のレシピを用いて、2,087の材料を100次元のベクトルにembeddingしたもの (最終的にはT-SNEで2次元に落としている)

image

word2vecでは、(king)(man)+(woman)=(queen)(king) – (man) + (woman) = (queen)という式がそれぞれの単語を表すベクトル同士の関係として成立したというのが有名だが、今回のfood2vecでも同じような式が成り立ったという.

(bacon)(egg)+(orangejuice)=(coffee)(bacon) – (egg) + (orangejuice) = (coffee)

(butter)(bread)+(roastbeef)=(sage)(butter) – (bread) + (roast beef) = (sage)

(なんだかわかったようなわからないような…)

Links