When and how should humans intervene? Trade-offs among time, interpretability, and accuracy in housing engagement data using AI
Date:
Large language models can read public comments at scale, but a credible result still takes human work — and it is rarely clear how much, or where it pays off. Using the public record of 350 Vancouver rezoning projects, this paper prices each stage of human input against the accuracy it buys, across topic classification, sentiment, and named entity recognition.
See the working project for the full description.