Fumiko Kano Glückstad

Unsupervised Knowledge Structuring: Application of Infinite Relational Models to the FCA Visualization

This work presents a conceptual framework for learning an ontological structure of domain knowledge, which combines Jaccard similarity coefficient with the Infinite Relational Model (IRM) by (Kemp et al. 2006) and its extended model, i.e. the normal-Infinite Relational Model (n-IRM) by (Herlau et al. 2012). The proposed approach is applied to a dataset where legal concepts related to the Japanese educational system are defined by the Japanese authorities according to the International Standard Classification of Education (ISCED). Results indicate that the proposed approach effectively structures features for defining groups of concepts in several levels (i.e., concept, category, abstract category levels) from which an ontological structure is systematically visualized as a lattice graph based on the Formal Concept Analysis (FCA) by (Ganter and Wille 1997).

Application of an Automatic Data Alignment & Structuring System for Intercultural Consumer Segmentation Analysis

This position paper introduces a conceptual framework of our ambitious international research project where the aim is extraction and alignment of heterogeneous consumer segment structures across a multiplicity of markets and cultures. We argue that an automatic data alignment and structuring system employing a non-parametric Bayesian relational modelling is an ideal approach that can address challenges in the conventional cross-cultural data analysis. The paper presents an example of our preliminary work that applies this approach to the analysis of opinion survey responses given by male populations in Sweden and Japan. The framework successfully extracts groups of males who express similar but also dissimilar response patterns from the two selected countries. Based on these preliminary studies, the paper discusses potential contributions and future challenges of the international consumer analysis project.

Nonparametric statistical structuring of knowledge systems using binary feature matches

Structuring knowledge systems with binary features is often based on imposing a similarity measure and clustering objects according to this similarity. Unfortunately, such analyses can be heavily influenced by the choice of similarity measure. Furthermore, it is unclear at which level clusters have statistical support and how this approach generalizes to the structuring and alignment of knowledge systems. We propose a non-parametric Bayesian generative model for structuring binary feature data that does not depend on a specific choice of similarity measure. We jointly model all combinations of binary matches and structure the data into groups at the level in which they have statistical support. The model naturally extends to structuring and aligning an arbitrary number of systems. We analyze three datasets on educational concepts and their features and demonstrate how the proposed model can both be used to structure each system separately or to jointly align two or more systems. The proposed method forms a promising new framework for the statistical modeling and alignment of structure across an arbitrary number of systems.

Analysis of Conceptualization Patterns across Groups of People

This paper analyzes patterns of conceptualizations possessed by different groups of subjects. The eventual goal of this work is to dynamically learn and structure semantic representations for groups of people sharing domain knowledge. In this paper, we conduct a survey for collecting data representing semantic representations of 34 subjects with different profiles in gender and educational background. The collected data is analyzed by an approach combining two extended versions of the Infinite Relational Model (Kemp et al. 2006): multiarray Infinite Relational Model (Mørup et al. 2010) and normal Infinite Relational Model (Herlau et al. 2012). Results indicate that the employed approach not only localizes similar patterns of conceptualization within a group of subjects having a common profile, but also identifies differences in conceptualization across different subject groups.

Examination of Heterogeneous Societies: Identifying Subpopulations by Contrasting Cultures

The recent development of data analytic tools rooted around the Multi-Group Latent Class Analysis (MGLCA) has enabled the examination of heterogeneous datasets in a cross-cultural context. Although the MGLCA is considered as an established and popular cross-cultural data analysis approach, the infinite relational model (IRM) is a new and disruptive type of unsupervised clustering approach that has been developed recently by cognitive psychologists and computer scientists. In this article, an extended version of the IRM coined the multinominal IRM—or mIRM in short—is applied to a cross-cultural analysis of survey data available from the World Value Survey organization. Specifically, the present work analyzes response patterns of the Portrait Value Questionnaire (PVQ) representing Schwartz’s 10 basic values of Japanese and Swedes. The applied model exposes heterogeneous structures of the two societies consisting of fine-grained response patterns expressed by the respective subpopulations and extracts latent typological structures contrasting and highlighting similarities and differences between these two societies. In the final section, we discuss similarities and differences identified between the MGLCA and the mIRM approaches, which indicate potential applications and contributions of the mIRM and the general IRM framework for future cross-cultural data analyses.

Alignment of Remote Cultures: In Contrast to the Relevance Theory of Communication

As the role of ontology in a multilingual setting becomes important to Semantic Web development, it becomes necessary to understand and model how an original conceptual meaning of a Source Language word is conveyed into a Target Language translation. Terminological ontology [1] is a tool used for knowledge sharing and domain-specific translation, and could potentially be suitable for simulating the cognitive models explaining real-world inter-cultural communication scenarios. In this paper, a framework referred to as the Relevance Theory of Communication [2] is contrasted to an empirical study applying Tversky´s contrast model [3] to data-sets obtained from the terminological ontology. The results indicate that the alignment of two language-dependent terminological ontologies is a potential method for optimizing the relevance required in inter-cultural communication, in other words, for identifying corresponding concepts existing in two remote cultures.