@inproceedings{oai:ipsj.ixsq.nii.ac.jp:00194770,
 author = {梅本, 晴弥 and 豊田, 哲也 and 大原, 剛三 and Haruya, Umemoto and Tetsuya, Toyota and Kouzou, Ohara},
 book = {行動変容と社会システム vol.05},
 month = {Mar},
 note = {近年，インターネット上にて投稿・共有されている大量の料理レシピを健康の管理もしくは増進に活用するためには，各料理に含まれる栄養素の量を計算する必要がある．
しかし，一般ユーザが投稿する料理レシピには食材名の表記揺れが多分に存在するため，栄養素量を自動で計算することは容易ではない．
そこで本研究では，料理レシピのタイトルから容易に抽出可能な料理カテゴリを料理レシピ中の単語に対する分散表現から予測するモデルを構築・学習し，その部分構造とし得られる食材名エンコーダを利用することで，食材名の表記揺れに頑健な栄養素推定法を実現する．
評価実験では，単語の分散表現を直接利用する手法および編集距離を利用する手法との比較を通して，提案手法の有用性を示す．, Recently, a huge number of cooking recipes have been posted and shared on the Internet. To utilize those recipes for healthcare or health promotion, it is necessary to correctly compute nutrients contained in dishes made according to them. However, there exit a lot of variations of an ingredient name in cooking recipes posted by anonymous users, which prevents us from automatically compute nutrients in a dish. To overcome this problem, in this paper, we construct a food name encoder which outputs a distributed representation of an ingredient name. To this end, we first solve a problem of estimating the category of a given cooking recipe with a neural network model that takes distributed representations of words in a cooking recipe. We can extract the targeted encoder as its sub-network of the neural network model learned. Distributed representations of ingredient names provided by the encoder enable us to estimate nutrients from cooking recipes more correctly, which is robust to the variations of ingredient names. Through experimental evaluations, we show the advantage of the proposed method over both existing ones that use edit distances between ingredient names and ones that directly use distributed representations of words in recipes.},
 publisher = {情報処理学会},
 title = {食材名の分散表現学習を用いた料理レシピの栄養推定手法},
 volume = {2019},
 year = {2019}
}