MiDATA and UTMD Announce Joint Research Results Using Matching App 'CoupLink'

MiDATA, an AI and data analysis consulting firm, announced the results of joint research with the University of Tokyo Market Design Center (UTMD). UTMD proposed a method to mitigate 'popularity concentration (congestion),' a structural issue in two-sided platforms, and improve fairness and matching quality, which MiDATA then implemented as a new recommendation algorithm. This algorithm was pilot-tested on the matching app 'CoupLink,' operated by Linkbal Inc., a shareholder of MiDATA, and its effectiveness was confirmed. The research paper is available on the preprint server 'arXiv' and the UTMD website. Since July 2024, MiDATA and UTMD have been researching 'two-sided recommendation' systems that consider the preferences and behavioral probabilities of both parties. In two-sided platforms like matching apps, where matches only occur when users mutually like each other, traditional methods tend to cause 'congestion,' where popular users receive excessive recommendations and 'likes.' This leads to inefficiencies such as popular users being overwhelmed and general users being buried. To address this, they combined UTMD's cutting-edge knowledge in 'market design (matching theory)' with MiDATA's 'advanced AI implementation capabilities' to develop the 'ECDA (Exposure-Constrained Deferred Acceptance)' algorithm. This algorithm applies the 'Deferred Acceptance algorithm,' which was the subject of a Nobel Prize in Economics, to the recommendation system. By setting appropriate upper limits on the number of recommendations (exposure) for individual users based on AI-predicted 'expected number of likes or matches,' it systematically mitigates excessive recommendations for specific users. The field test confirmed that the extreme concentration on the top 0.1% was corrected, leading to improved matching opportunities that facilitate actual interaction, fairer meeting opportunities, and a safer environment. Moving forward, they plan to apply these insights to data analysis consulting services and provide this technology and DX support to platform operators in other industries facing similar 'opportunity loss due to popularity concentration,' such as job sites and freelance matching services.