This paper investigates how the Infinite Relational Model (IRM) [Kemp 2006], a novel unsupervised machine learning method, is effectively applied to loosely-structured datasets consisting of concepts and features for the purpose of mapping Culturally Specific Concepts (CSCs) in a multi-cultural context. The aim of this investigation is two-fold: i) to identify an effective strategy of applying the IRM for the purpose of CSC-mapping; and ii) to investigate possibilities of applying the IRM for efficiently constructing feature-based ontologies that are multi-culturally interoperable. Accordingly, three strategies are tested in our experiments: 1) applying the IRM directly to two CSC-feature matrices, respectively representing the educational domain knowledge in Japan and Denmark for first categorizing them into categorical classes that are to be subsequently compared and aligned; 2) applying the IRM directly to a matrix where the two CSC-feature matrices respectively representing the Danish- and Japanese educational domain knowledge are merged; and 3) applying the Bayesian Model of Generalization (BMG) [Tenenbaum 2001] to directly compute similarity relations between CSCs in the two cultures at hand, thereafter to apply the IRM for clustering CSCs in the respective cultures into categorical classes. The results indicate that the third strategy seems to be the most effective approach for not only clustering CSCs into more specific and appropriate categorical classes but also for capturing complex relationships between each categorical classes existing in the two cultures.


Antal sider10
StatusUdgivet - 2012
Scopus citationer
BegivenhedThe 26th Annual Conference of The Japanese Society for Artificial Intelligence. 2012 - Yamaguchi, Japan


KonferenceThe 26th Annual Conference of The Japanese Society for Artificial Intelligence. 2012
SponsorThe Japanese Society for Artificial Intelligence

ID: 35602972