Proposal of Finding Potentially Valid Menus from Recipe Dataset Using Knowledge Graph Embedding

This paper proposes a method to find potentially valid menus which do not explicitly exist in the recipe dataset using knowledge graph embedding (KGE).KGE can predict the missing links in Knowledge Graphs (KGs) by embedding each entity and relation as a vector. Using this feature,the proposed method finds potentially valid menus from the recipe dataset. As it is impossible for a recipe KG to include all possible combinations of recipes that could be regarded as menus,finding potentially valid menus is necessary for realizing recipe recommender systems.This paper describes how to construct the recipe KG from the Cookpad dataset and find potentially valid menus by exploiting TransE.The effectiveness of the proposed method is shown based on the survey-based evaluation. Furthermore, this paper investigates the effect of additional learning.