Associate Professor Matsushima, a collaborating researcher, gave a poster presentation at the SICE Annual Conference on Systems and Information Engineering (SSI2026), organized by the Society of Instrument and Control Engineers.

Presentation Information
Analysis of Dataset Selection Behavior on Kaggle toward Building an Agent-Based Model
Hiroyasu Matsushima (Kanazawa University), Teruaki Hayashi (Hitotsubashi University), Takumi Shimizu (Keio University)

Data marketplaces, where data providers and users exchange data, remain at an early stage in which various platforms continue to emerge, and knowledge about institutional designs that can sustain data provision and use over the long term is still limited. This study is part of a broader effort to clarify, through agent-based simulation, how such institutions and the behavior of participants affect the structure and dynamics of the market as a whole.

As a first step toward building a new simulation model, this presentation analyzed the criteria users apply when selecting a new dataset for their analysis.

Using a conditional logit model, the study estimated the relative likelihood of selection by comparing candidate datasets available at the same selection opportunity. The results showed that datasets with a greater track record of use and of votes were more likely to be selected, while datasets were less likely to be selected the longer the time since their publication. This pattern held consistently even when the number of candidate datasets considered was varied.

Going forward, the research will examine the conditions under which dataset publication and user engagement continue over time, with the aim of building an agent-based simulation model of data marketplaces.