Concurrence of Word Concepts in Cooking Recipe Search

Michiko Yasukawa, Falk Scholer

Abstract: Recipe search systems rely on keyword matching, and in this work we analyze the consistency of vocabulary, investigating how agreement differs when searching: for ingredients versus dishes; for common versus uncommon items; and, between recipe authors and searchers. The experiments for this study use a crowd-sourcing framework and a large corpus of cooking recipes. User data is compared with author data from a large collection of written cooking recipes. Our results show that (i) names for ingredients are more concurrent than for dishes; (ii) concurrency is higher for common items than uncommon items; and (iii) inadequate word representation in the recipe corpus leads to poor concurrence of vocabulary between searchers and authors. These results suggest fundamental knowledge to improve the effectiveness of cooking recipe search.