Closing the Welfare Outreach Gap: A Conversational Architecture and Cell-Level Eligibility Benchmark for Korean Welfare Recommendation

Byeongmin Kang, Minwoo Han, Junhak Lee, Minsu Kim, Jihie Kim

Keywords: AI for Good, conversational recommendation, welfare policy recommendation, eligibility benchmark, Korean public sector, outreach gap, cell-level evaluation, conversational elicitation, knowledge resources, public-sector AI

Abstract: Welfare benefits often fail to reach the people who need them most—elderly citizens, digitally underserved users, and those unfamiliar with eligibility categories—because portal-based search requires users to know both which programs to search for and how their personal attributes map to eligibility criteria. This paper addresses this structural welfare outreach gap in three steps: (1) we propose a conversational welfare-policy recommendation architecture that elicits user attributes through natural-language dialogue; (2) for quantitative comparative evaluation of its core eligibility-matching component, we construct KWelfareBench, a cell-level eligibility ground-truth table over 4,937 Korean welfare policies and 180 synthetic personas; and (3) using this benchmark, we compare multiple recommendation architectures and identify an eligibility-matching structure suited to the Korean welfare domain. Together, these contributions provide an empirical foundation for designing recommendation systems that reliably connect users to eligible welfare policies through dialogue—even when users lack familiarity with eligibility categories—and the resources, code, and personas are released as a common substrate for subsequent Korean welfare outreach research.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 409