Construction of metadata database structured by conceptual elements of text structure and semantic search evaluation of Korean studies

Abstract

Purpose This study aims to develop metadata of conceptual elements based on the text structure of research articles on Korean studies, to propose a search algorithm that reflects the combination of semantically relevant data in accordance with the search intention of research paper and to examine the algorithm whether there is a difference in the intention-based search results. Design/methodology/approach This study constructed a metadata database of 5,007 research articles on Korean studies arranged by conceptual elements of text structure and developed F1(w)-score weighted to conceptual elements based on the F1-score and the number of data points from each element. This study evaluated the algorithm by comparing search results of the F1(w)-score algorithm with those of the Term Frequency- Inverse Document Frequency (TF-IDF) algorithm and simple keyword search. Findings The authors find that the higher the F1(w)-score, the closer the semantic relevance of search intention. Furthermore, F1(w)-score generated search results were more closely related to the search intention than those of TF-IDF and simple keyword search. Research limitations/implications Even though the F1(w)-score was developed in this study to evaluate the search results of metadata database structured by conceptual elements of text structure of Korean studies, the algorithm can be used as a tool for searching the database which is a tuning process of weighting required. Practical implications A metadata database based on text structure and a search method based on weights of metadata elements – F1(w)-score – can be useful for interdisciplinary studies, especially for semantic search in regional studies. Originality/value This paper presents a methodology for supporting IR using F1(w)-score—a novel model for weighting metadata elements based on text structure. The F1(w)-score-based search results show the combination of semantically relevant data, which are otherwise difficult to search for using similarity of search words.