An Automatic Framework Recognizing the Relationships of Cultural Heritage

Abstract

The field of cultural heritage has developed over a long period, accumulating a wealth of research findings. Researchers are now focusing on systematic relationships and taxonomic studies of heritage, exploring the underlying cultural information embedded within. Inspired by the fields of machine learning and biology, we propose a research approach that combines data processing with unsupervised algorithms (“Feature Sparsity Module + N”), which can be utilized to unveil the systematic relationships of cultural heritage study subjects. We construct the Cultural Heritage Relationship Evaluation Framework (CHREF), framework using the structure of “FSM + PCA + HCA”, which offers a workflow characterized by interpretability, visualization capabilities, and automation. The framework utilizes the FSM module to transform the research subjects into matrices, employing PCA and HCA to obtain intuitive charts and reliable data results with minimal manual intervention. Additionally, we provide experiments and a user study on traditional Chinese brick kilns to validate the effectiveness and universality of the proposed framework. The “FSM + N” methodology and CHREF can provide tools for various stages of work in cultural heritage, making significant contributions to the digital development and database construction in the field of cultural heritage.

Keywords

• Cultural heritage
• machine learning
• interdisciplinary research
• embedding
• one-hot encoding
• principal component analysis