An Automated Approach for Information and Referral of Social Services Using Machine Learning

Abstract

The Information and Referral Federation of Los Angeles County (211 LA County) is a nationally recognized service center that makes referrals to those in need of social service resources available at sites throughout Los Angeles County and nationally for those in need and for at-risk populations. Referrals are currently made using an on-line web-based referral system backed by a rich highly curated dataset collected over years and informed by a national taxonomy of social services. In support of resource referrals both for our on-line system, and for a new website presence, our research team has investigated and realized an automated resource referral system that learns from a caller's demographic information and historical referral data collected by human experts to recommend sites at the time of an active call. This system leverages a state of art multi-label neural network classifier, tuned by grid search for obtaining the best hyper-parameters for this system. The automated approach we have created allows 211 LA County to interactively provide a meaningful referral to those in need. In this paper, we describe our evaluation strategy and accuracy of our system on a one-year dataset containing over 450 thousand calls.