Building k-partite association graphs for finding recommendation patterns from questionnaire data

Graph-pattern association rules have been explored for detecting frequent subgraph structures in real-world network data, which can reveal new insights for decision-making, recommender systems, and predictive models. However, questionnaire data have been neglected so far even though they are one of the most affordable ways to gather quantitative data. Questionnaires can cover every aspect of a topic, generating new strategies and trends for many organisations. The challenge is twofold: develop a model for handling nominal/Boolean data and ordinal data simultaneously, as well as multiple values assigned to a single item. In this article, the synergy between the well-known Apriori algorithm and k-partite graph modelling is proposed to discover frequent recommendation patterns from questionnaire data. Using graph centrality and similarity measures, the large number of association rules are further analysed to discover meaningful spatial structures in non-metric spaces. Counting triangles is also used to uncover hidden thematic structures of link recommendations. We demonstrate how our proposed approach can be applied to a tourism questionnaire survey to reveal frequent patterns in k-partite graphs, which can further be used for recommender systems.