Information Extraction for Cultural Heritage Knowledge Acquisition Using Word Vector Representation

Abstract

Cultural heritage is the legacy of social values, traditions and beliefs a group of people inherited from past generation. It can give people to understand the history better where a specific group of people come from and why they behave in their way of life. To preserve cultural heritage, information extraction acquires textual information from various data sources and then converts the information extracted to structured representations. This cultural information is kept digitally for future generations. However, since many approaches on information extraction focus on determining name entities and relationships between a pair of entities, the lack of explicit semantic concept relations makes it difficult for these approaches to apply in a variety of applications. This paper introduces an approach to extracting information from various cultural heritage data sources using word embedding with feature extraction. We focus on identifying named entities and determining the semantic relation triple such as part of speech tags and position tags by using conditional probability for discovering this triple relationship. The method was evaluated using CIDOC-CRM, a common standard ontology for cultural heritage information. The results demonstrated that our approach can achieve high accuracy at 81%.