Harnessing Large Language Models for Efficient Crowd Management in Large-Scale Events
Eventour is an open-source framework that leverages Large Language Models (LLMs) and generative AI to help event organizers reduce congestion and improve attendee experience by:
- Building an enriched knowledge graph of urban Points of Interest (POIs), transportation infrastructure, environmental data, and more
- Generating personalized, context-aware walking itineraries that guide attendees through cultural landmarks before dispersing them toward metro stations
- Incorporating gamification elements (interactive quizzes and puzzles) to engage users and encourage staggered departures, mitigating peak‐time crowding
- Providing a modular codebase for data ingestion, preprocessing/enrichment, distance matrix computation, LLM fine-tuning, itinerary generation, and gamification content
- Offering a React Native mobile application prototype (“Eventour”) for on-the-go delivery of itineraries and quizzes
-
Data Integration & Enrichment
– Ingests POIs from Wikidata, municipal open datasets (trees, benches, fountains, historic shops, NIL zones), and real-time weather
– Cleans, normalizes, de-duplicates, geocodes, and semantically tags records for a harmonized spatial knowledge graph -
Distance Matrix Computation
– Uses OpenStreetMap data and OSRM to compute pedestrian distances and durations via a Dockerized routing engine -
LLM-Powered Itinerary Planning
– Fine-tunes Mistral-7B with LoRA on abstracted distance matrices to learn itinerary rules (continuity, transport end, no duplicates, bidirectionality)
– Generates multiple staggered itineraries ending at metro stations, with configurable number of stops and intervals -
Generative Gamification
– Retrieves POI metadata from Wikidata/DBpedia and uses Mistral-7B to produce multiple-choice questions and fun facts
– Integrates quizzes into the walking experience to sustain engagement -
Prototype Mobile App
– React Native + Expo codebase for delivering itineraries and quizzes on iOS and Android