Using face detection in photographs and cluster analysis to support exploration of social relationships between historical personages in a biographical database

• Main Authors: Sie, Shun-Hong, Ke, Hao-Ren, Chang, Su-Bing
• Other Authors: National Taiwan Normal University
• Format: Article
• Language: English
• Published: 2022
• Subjects: Library and information science
• Online Access: https://hdl.handle.net/10356/154748
• Institution: Nanyang Technological University

Background. The Taiwan Biographical Database (TBDB) assembles biographical information of historical personages in Taiwan. It is a digital-humanities-oriented system that supports relational database operations, fulltext search, social network analysis, and geographic information system functions. Objectives.Through semi-automatic named entity recognition from the fulltext of biographies, TBDB assists historians to construct networks of social relationships. However, the fulltext of biographies may not describe all social relationships. Taking advantage of the fact that historical photographs were usually taken on formal occasions, historical photographs may be exploited to uncover additional relationships. This paper describes and evaluates a face detection function in TBDB that utilizes the OpenCV Library to detect faces of historical persons in old photographs. Furthermore, it employs hierarchical agglomerative clustering to combine fragmentary social networks. Results. An experiment using 45 historical photographs found that the face detection function achieved an average recall of 98% recall, but with low precision. To address the low precision rate, a user interface has been implemented in TBDB to facilitate review and deletion of false-positive faces in the photographs. Furthermore, cluster analysis is used to integrate social relationships found in biographies, those detected from historical photographs, and even relationships harvested from external sources, to produce comprehensive social networks for historical research.