Construction of Makeup Vocabulary Datasets for Scene Search on Makeup Movies

Abstract

This study aims to acquire the vocabulary of cosmetics and makeup techniques used for searching for an intended scene from makeup movies. When applying cosmetics, it is necessary to appropriately select a set of cosmetics by considering individual preferences, affective senses, and physical constitutions. Makeup movies are gaining popularity as reference information for applying makeup and are becoming as crucial as traditional information media. Currently, people retrieve makeup videos on a video-by-video basis, and methods for retrieving specific processes in makeup movies have not yet been developed. As the first step in developing such a system, this paper conducted a fundamental analysis of vocabulary collected from cosmetic reviews and speech in makeup movies. The characteristics of each modal were studied, focusing on the semantics of sentences and word categories. As a result, each data source contains different types of cosmetic vocabulary. This result implies that using each data source in a combining manner to build a vocabulary is a rational way to expand the vocabulary efficiently.