LOCALYN:A Hyperlocal Multi-Source Civic Event Intelligence Platform

Uncategorized

Authors: Roopa T, Ayush Gowda P, Kumar K C, Tarun C L

Abstract: Information about road conditions, public infrastructure and local disruptions is recorded by several city-facing organisations. Since each organisation follows its own categories and reporting practices, observations about one incident may remain scattered across separate systems. This paper presents LOCALYN, a hyperlocal event-intelligence platform developed around Bengaluru civic data. The implemented system combines public-source ingestion with a cross-agency association module that links Bruhat Bengaluru Mahanagara Palike (BBMP) grievance categories to Bengaluru Traffic Police (BTP) incident causes. Matching uses a curated administrative crosswalk, ward compatibility, timestamp comparison and text similarity. A four-feature score retains the contribution of each signal, using the all-MiniLM-L6-v2 sentence-embedding model when available and Jaccard similarity as a fallback. The geographic service processes 225 ward polygons through Shapely and an STRtree index, with explicit handling for polygon boundaries and nearby-centroid fallback. Accepted record links are grouped into canonical events, which receive an interpretable Confidence from Heterogeneous Evidence Fusion (CHEF) score. The wider application includes news, BBMP and weather ingestion modules, SQLite storage, FastAPI endpoints and a React–Leaflet interface. In the implemented nine-pair benchmark, LOCALYN obtains an F1-score of 1.00; the date-and-ward baseline obtains 0.833, while the date–ward–category baseline obtains 0.000. These results provide an initial functional comparison of the implemented matching workflow. The paper describes the system architecture, matching procedure and current evaluation, and identifies extensions for live agency integration and larger-scale testing.

DOI: https://doi.org/10.5281/zenodo.24394104

× How can I help you?