Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PUBLICSPEAK: Hearing the Public with a Probabilistic Framework in Local Government

Tianliang Xu, Eva Maxfield Brown, Dustin Dwyer, Sabina Tomkins

TL;DR

PUBLICSPEAK addresses the challenge of extracting public remarks from local government meetings by fusing meeting structure with linguistic cues through Probabilistic Soft Logic (PSL). The approach jointly infers meeting sections, speaker roles, and remark types using predicates such as Section$(M,U,X)$, SpeakerRole$(M,S,Y)$, and Remarktype$(M,U,T)$ within a hinge-loss Markov random field, enabling robust public-remark discovery. Evaluated on ~30K remarks from seven cities, PublicSpeak outperforms state-of-the-art baselines by up to 40% and demonstrates strong generalization to unseen cities via Leave-One-City-Out experiments. The work further delivers an end-to-end pipeline for generating open datasets of public remarks, analyzes public-topic dynamics across cities, and situates AI-assisted journalism as a tool for transparency rather than automation of reporting.

Abstract

Local governments around the world are making consequential decisions on behalf of their constituents, and these constituents are responding with requests, advice, and assessments of their officials at public meetings. So many small meetings cannot be covered by traditional newsrooms at scale. We propose PUBLICSPEAK, a probabilistic framework which can utilize meeting structure, domain knowledge, and linguistic information to discover public remarks in local government meetings. We then use our approach to inspect the issues raised by constituents in 7 cities across the United States. We evaluate our approach on a novel dataset of local government meetings and find that PUBLICSPEAK improves over state-of-the-art by 10% on average, and by up to 40%.

PUBLICSPEAK: Hearing the Public with a Probabilistic Framework in Local Government

TL;DR

PUBLICSPEAK addresses the challenge of extracting public remarks from local government meetings by fusing meeting structure with linguistic cues through Probabilistic Soft Logic (PSL). The approach jointly infers meeting sections, speaker roles, and remark types using predicates such as Section, SpeakerRole, and Remarktype within a hinge-loss Markov random field, enabling robust public-remark discovery. Evaluated on ~30K remarks from seven cities, PublicSpeak outperforms state-of-the-art baselines by up to 40% and demonstrates strong generalization to unseen cities via Leave-One-City-Out experiments. The work further delivers an end-to-end pipeline for generating open datasets of public remarks, analyzes public-topic dynamics across cities, and situates AI-assisted journalism as a tool for transparency rather than automation of reporting.

Abstract

Local governments around the world are making consequential decisions on behalf of their constituents, and these constituents are responding with requests, advice, and assessments of their officials at public meetings. So many small meetings cannot be covered by traditional newsrooms at scale. We propose PUBLICSPEAK, a probabilistic framework which can utilize meeting structure, domain knowledge, and linguistic information to discover public remarks in local government meetings. We then use our approach to inspect the issues raised by constituents in 7 cities across the United States. We evaluate our approach on a novel dataset of local government meetings and find that PUBLICSPEAK improves over state-of-the-art by 10% on average, and by up to 40%.

Paper Structure

This paper contains 15 sections, 2 equations, 1 figure, 8 tables.

Table of Contents

  1. Introduction
  2. Hearing From the Public, a Data-Driven Opportunity
  3. Public Remarks
  4. Problem Definition
  5. Our Approach
  6. Probabilistic Soft Logic
  7. PublicSpeak
  8. Empirical Evaluation
  9. Dataset
  10. Quantitative Evaluation
  11. Training Details
  12. Generalizing to a New City
  13. Hearing the Public
  14. Discussion
  15. Contributions to Prior Work

Figures (1)

  • Figure 1: Topics assigned to the public comments of each city. The top-left figure shows the topic assignments across pre-processed public comments in the training set for each city ($\sim$1700 total), the top-right figure shows the topic assignments for those remarks in the test set which are known to be public comments ($\sim$350 total), and the bottom figure shows the topic assignments for those remarks in the test set which are predicted to be public comments by PublicSpeak ($\sim$350 total). We find that the top-right and bottom figures lead to qualitatively similar findings, indicating the efficacy of our approach.