BURST
NAWA Bekker 2025 · ISTI-CNR Pisa

BURST

Balanced Urban Recommenders for Sustainable Mobility Transitions

Google Maps, Uber and TripAdvisor quietly decide where millions of people go and how they get there. BURST asks what those algorithms do to a city and designs recommenders that keep people moving without concentrating everyone in the same streets.

  • 2026–2028
  • 24-month postdoctoral fellowship
  • Pisa, Italy
How it works
Spatial filter bubble versus balanced recommendation Two schematic city maps showing the same ten trips, starting from the same ten scattered origins. On the left, a conventional urban recommender sends every trip across the city into one of just two overloaded destinations. On the right, a balanced recommender distributes them across ten destinations spread evenly through the city, each close to where its trip began. Urban Recommender Balanced Urban Recommender 7 3
The same ten trips from the same ten origins. Left: every route ends in one of two recommended destinations. Right: ten destinations spread across the city.
The problem

Spatial filter bubbles

Urban Recommenders (URs) are the AI systems embedded in navigation, ride-hailing and review platforms and are designed for individual convenience. Because they steer everyone toward the same narrow set of popular places and routes, they concentrate activity, amplify popularity bias and reinforce a monocentric urban structure.

What happens

  • Traffic concentration, pollution and noise increase in a handful of neighbourhoods
  • Over-tourism and resident backlash at popular venues
  • Drop in footfall and economic activity in overlooked districts
  • Peri-urban shortcuts turn quiet lanes and cycling routes into through-traffic corridors

Why it is unstudied

Research on URs has focused on aggregate effects, such total congestion, emissions, and venue popularity. The individual-level formation of spatial filter bubbles, and their role in reinforcing spatial inequality, has not been systematically explored. Routing and destination recommendation are usually studied apart; BURST treats them as one system.

Today’s notion of “individual convenience” is also narrow: the fastest or shortest path, ignoring safety, pleasantness and broader urban wellbeing.

Cities are already reacting. For example, Barcelona removed a bus route from tourist apps to reduce overcrowding and protect local access. It is now urgent to understand how recommenders shape mobility and exposure, and how to mitigate it.
Research questions

Three questions, three objectives

Each research question maps onto a specific objective and a project phase.

RQ1 → SO1

Detect

How can we detect and quantify the presence and intensity of spatial filter bubbles in mobility trajectories? We will develop methods to detect and measure the intensity of spatial filter bubbles.

RQ2 → SO2

Measure

What are the long-term effects on congestion, socio-spatial segregation, environmental sustainability and the travel experience itself, including safety and perceived convenience? We will measure the impact of spatial filter bubbles in different settings.

RQ3 → SO3

Mitigate

Can we design recommenders that mitigate filter-bubble effects while aligning individual preferences with accessibility, equity and sustainability? This project will deliver a new class of Balanced Urban Recommenders (BURs).

Approach

State-of-the-art methodology

A modular, simulation-based methodology. High-resolution mobility data and open road and traffic datasets coupled with traffic microsimulations, UR simulations replicating commercial routing and destination strategies.

Phase 1

Detection & quantification

  • Collate high-resolution mobility, road and traffic datasets
  • Develop metrics for concentration, diversity and exposure
  • Empirically analyse individual and collective mobility patterns
Phase 2

Digital twins & simulation

  • Build data-driven mobility models for four case studies
  • Couple them with traffic microsimulations
  • Run A/B tests, scenario analysis and quasi-experiments
Phase 3

Balanced Urban Recommenders

  • Design BUR prototypes combining individual and collective goals
  • Benchmark against conventional popularity-driven URs
  • Evaluate on spatial equity, congestion, emissions and exposure

Tools

Existing open-source tools from both the fellow and the host group:

Work plan

Timeline

Five research tasks across the 24-month fellowship, September 2026 to August 2028.

  • Detection & metrics Sep 2026 – May 2027
  • Digital twins Mar 2027 – Sep 2027
  • Simulation experiments Sep 2027 – Mar 2028
  • BUR design & evaluation Nov 2027 – Aug 2028
  • Dissemination & open science Sep 2026 – Aug 2028
Project running
Sep 2026Sep 2027Aug 2028
Expected results

What BURST will deliver

Everything is released openly, following FAIR and open-science principles: code and framework on GitHub, case-study analyses and data on Zenodo with permanent identifiers, papers in open access.

  1. 01Four empirical case studies — two urban, two peri-urban — each evaluated with at least ten indicators covering equity, mobility, congestion, exposure and emissions at individual and systemic level.
  2. 02Two BUR prototypes implementing alternative recommendation logics — diversity promotion, environmental sustainability, spatial equity — benchmarked against standard URs.
  3. 03An open-source toolkit bundling the metrics, the simulation framework and the BUR prototypes, with documentation and example case studies.
  4. 04Three peer-reviewed articles on filter-bubble detection, long-term UR impacts, and BUR design and evaluation.
  5. 05A policy brief translating the findings into actionable guidance for urban authorities on UR impacts and justice-aware recommender design.

Project status. BURST started in September 2026 and is in its first phase. Code, datasets and publications will appear in this repository and on Zenodo as they are released. This page is updated as the project progresses.

Publications & news

Latest from the project

Papers, preprints, datasets, software releases and project updates will be listed here as BURST progresses.

Nothing published yet

The project started in September 2026 and is in its first phase. All outputs will be released openly — code on GitHub, datasets and case-study analyses on Zenodo with permanent identifiers, and papers in open access.

People

Who is doing this

A postdoctoral fellowship hosted by the Knowledge Discovery and Data Mining Laboratory at ISTI-CNR in Pisa, bringing together human mobility modelling and research on human–AI coevolution in cities.

Dr Kamil Smolak

Dr Kamil Smolak

Fellow & principal investigator

Assistant Professor at the Institute of Geodesy and Geoinformatics, Wrocław University of Environmental and Life Sciences. Works on human mobility modelling, simulation and algorithmic fairness; author of the HuMobi library and former PI of an NCN Preludium grant on bias in mobility predictions.

ORCID GitHub HuMobi

Dr Luca Pappalardo

Dr Luca Pappalardo

Host supervisor

Senior Researcher at ISTI-CNR and Professor of Geospatial Analytics at the University of Pisa. Works on human mobility, data-driven urban analytics and human–AI coevolution; creator of the scikit-mobility library.

ORCID KDD Lab scikit-mobility

Funding & context

Supported by

Polish National Agency for Academic Exchange (NAWA)

Bekker NAWA Programme 2025, project BPN/BEK/2025/1/00028. September 2026 – August 2028.

ISTI-CNR, Pisa

Institute of Information Science and Technologies “Alessandro Faedo”, National Research Council of Italy — Knowledge Discovery and Data Mining Laboratory.

Wrocław University of Environmental and Life Sciences

Institute of Geodesy and Geoinformatics.