Analysis Framework for Next-Items Recommendation using Local Process Model on a Pairwise Comparison Dataset
Frans Prathama, Wenny Franciska Senjaya, Bernardo Nugroho Yahya, Jei‐Zheng Wu · 2019
Consumers often engage with comparison and attractive recommendations before making a decision to purchase. An intelligence approaches such as recommender system can be applied in order to provide recommendations to consumers while comparing products. While research works in recommendation system focused on transactional data with single item (i.e., market basket), there are some challenges on pairwise comparison, i.e., multiple items at the same time and order sequence of items. In addition, the next-items recommendation is a challenge on a pairwise comparison data due to its characteristics; sparsity and intransitivity. The mentioned challenges can influence consumers’ decision during product search. To address the challenges, this study proposes a new framework by combining two different approaches, i.e., association rules and sequential pattern mining, to generate a recommendation on a pairwise comparison data. Using top-k association rules, the sparsity data problem could be overcome. The result from association rules is suitable for constructing the local process model, as a technique of process mining to find the frequent sequential patterns due to the intransitivity. The result of local process model gives reasonable insights as to the recommender system.