AirTOWN: A Privacy-Preserving Mobile App for Real-time Pollution-Aware POI Suggestion

Giuseppe Fasano, Yashar Deldjoo, Tommaso Di Noia · 2025

Read AirTOWN: A Privacy-Preserving Mobile App for Real-time Pollution-Aware POI Suggestion by Giuseppe Fasano, Yashar Deldjoo, Tommaso Di Noia (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-13608.txt.

Introduction

Giuseppe Fasano, Yashar Deldjoo, and Tommaso Di Noia present a demo app in a 2025 paper that suggests city places to visit while minding air quality. The app, called AirTOWN, joins live air quality data with each user's tastes. The aim is to help people choose spots that are healthier to visit at that moment. Personal suggestions come from collaborative filtering, a method that learns from users with like tastes.

Privacy is guarded by federated learning, so personal data need not be pooled in one place. Air data come from sensor networks in cities such as Bari in Italy and Cork in Ireland, named in the paper as the United Kingdom case. Where sensors are sparse, interpolation fills in values between them. The authors present the system as a health aware guide that adjusts to current city air while keeping user data safe.

Catalog record

CREATORGiuseppe Fasano, Yashar Deldjoo, Tommaso Di Noia
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE9 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-13608

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

AirTOWN: A Privacy-Preserving Mobile App for Real-time Pollution-Aware POI Suggestion
Giuseppe Fasano, Yashar Deldjoo, Tommaso Di Noia

Abstract
This demo paper presents AirTOWN , a privacy-preserving mobile application that provides real-time, pollution-aware recommendations for points of interest (POIs) in urban environments. By combining real-time Air Quality Index (AQI) data with user preferences, the proposed system aims to help users make health-conscious decisions about the locations they visit. The application utilizes collaborative filtering for personalized suggestions, and federated learning for privacy protection, and integrates AQI data from sensor networks in cities such as Bari, Italy, and Cork, UK. In areas with sparse sensor coverage, interpolation techniques approximate AQI values, ensuring broad applicability. This system offers a poromsing, health-oriented POI recommendation solution that adapts dynamically to current urban air quality conditions while safeguarding user privacy.



1
Introduction

Air pollution is a global challenge, with 99% of the population breathing air exceeding WHO guideline limits
[
1
]
. This issue is particularly acute in urban areas due to vehicle emissions and industrial processes. Point-of-interest (POI) recommendation systems offer a unique opportunity to enhance user experiences and support public health by incorporating health-conscious suggestions.

This paper introduces
AirTOWN
, a mobile application that integrates real-time pollution data with user preferences to provide tailored POI recommendations for health-conscious urban navigation. The system emphasizes
personalization
,
pollution awareness
,
real-time sensory data integration
,
privacy preservation
, and considers the
distance
to venues.
AirTOWN
helps users make healthier choices by identifying nearby POIs, personalizing suggestions using collaborative filtering (CF), and re-ranking them based on Air Quality Index (AQI) data from AirSence
1
1

1

https://airsence.com/
.

To ensure accuracy,
AirTOWN
uses sensor networks in Bari, Italy, and Cork, UK, providing AQI, temperature, humidity, and pressure data. These cities represent diverse urban settings with varying air quality challenges. In areas with limited sensor coverage, interpolation techniques approximate AQI, extending the system’s applicability.

Existing systems often monitor exposure
[
14
]
or predict pollution along routes without suggesting healthier POIs
[
10
]
. Health-based recommendation systems focus on static health metrics
[
12
,
11
]
, and privacy-centric solutions
[
13
,
3
]
lack real-time environmental integration. In contrast,
AirTOWN
combines real-time pollutant data, personalization, and privacy to offer dynamic, health-conscious recommendations. While there are limitations,
AirTOWN
represents a significant step toward supporting healthier urban navigation through real-time data integration and privacy-conscious design.

Figure 1
:
Recommendation process in
AirTOWN

2
System Architecture

The architecture of
AirTOWN
is designed to leverage a client-server model featuring four layers: Application, Service, Interface, and Data Resources Layer.

Application Layer.
This layer serves as the user interface on mobile smartphones and performs POI suggestions integrating air quality information. To obtain an effective POI recommendation system (RS), we focused on balancing three key factors: personalization, pollution levels, and distance of the target venue (POI). When a user requests suggestions, the system first identifies POIs within a certain radius (e.g., 1 km) of the user’s current location, addressing the distance factor. Next, a collaborative filtering (CF) model predicts user preferences to personalize the recommendations. We chose the Matrix Factorization (MF) solution over more sophisticated alternatives as Deep Learning, because of its better transparency and lower complexity, allowing for distributed training on devices with limited resources. Finally, the system applies a weighted re-ranking function that balances the predicted preferences and the AQI of the POIs. The parameter
\alpha\in[0,1]
modulates the strength of AQI influence in the re-ranking process
S=\alpha\cdot S_{MF}+(1-\alpha)\cdot S_{AQI}
. To enhance MF performances, the model is trained on user preferences, which are locally collected via surveys on already visited POIs. The Application Layer never shares user data and the update of the local model is performed in a client-server implementation of FL paradigm, allowing for user privacy.

Service Layer.
Hosted on a server, the Service Layer manages back-end processes such as login, registration, and data retrieval for the Application Layer. It approximates AQI data for unsensed locations using radial basis function interpolation and performs federated learning rounds for global model training via the Federated Averaging protocol. Only item embedding updates are shared, while user embeddings remain local.

Interface and Data Resources Layer.
This layer comprises APIs linking the Service Layer with data resources. The AirSENCE API provides localized air quality data, while Directions and Places APIs enable navigation and location information. The Google Database manages Google-related data (Photos, place information, routing data), and the AirSENCE Database stores air quality measurements and federated model parameters.

3
Intra-user and Inter-user Demonstration

To evaluate the effectiveness of
AirTOWN
in providing personalized, health-conscious recommendations, we conducted intra-user and inter-user demonstrations in Bari, Italy, using synthetic AQI data due to limited sensor coverage. This setup enabled us to simulate varying air quality scenarios and assess the impact of recommendation parameters.
2
2

2

https://anonymous.4open.science/r/Airtown-Application/

Two users were simulated in Aldo Moro Square, Bari, each using
AirTOWN
on mobile devices. User preferences were collected to compute individual embeddings through local models, allowing for personalized recommendations. A virtual sensor grid structured as a
1\times 1
km layout surrounding each user provided AQI values randomly assigned within a range of 20 to 70, enabling realistic air quality variation. Both users requested restaurant recommendations, with User 1 representing a generally healthy individual and User 2 representing an elderly individual with increased sensitivity to air quality.

3.1
Demonstration Results

The demonstration involved adjusting the weighting parameter
\alpha
to control the influence of AQI versus user preferences in the recommendation ranking.

Intra-user demonstration.
As illustrated by the red-boxed results in Figure
2
, varying
\alpha
shows different recommendation lists:

•

\alpha=0
(AQI-driven):
Recommendations are ordered solely by AQI, disregarding preferences.

•

\alpha=1
(preference-driven):
Recommendations prioritize user preferences, sometimes listing POIs with higher AQI.

•

\alpha=0.5
(balanced):
Recommendations are adjusted to reflect both AQI levels and personal tastes, achieving a compromise between health considerations and preferences.

Inter-user demonstration.
User 2, representing an elderly individual, applied
\alpha=0.3
to favor lower AQI levels due to increased health sensitivity, as shown in the blue-boxed list in Figure
2
. This setting provided a recommendation list distinct from that of User 1, who prioritized preferences over AQI, demonstrating
AirTOWN
’s adaptability to individual health needs.

Figure 2
:
Recommendation lists for two users in the simulated scenario.

4
Conclusion

We presented
AirTOWN
, a mobile application that combines personalization, real-time pollution awareness, privacy preservation, and proximity considerations to deliver health-conscious point-of-interest (POI) recommendations. Initial experiments highlight the effectiveness of the Mobile App. in balancing user preferences with real-time air quality data, making it a valuable tool for urban navigation that promotes healthier choices.
Future work will focus on the privacy aspect, involving solutions to protect the updates, such as differential privacy. Further evaluation of the App. effectiveness will be conducted with a larger user base.

Overall, this paper situates
AirTOWN
within the broader context of trustworthy recommender systems
[
6
,
9
,
4
,
5
,
15
]
by emphasizing privacy preservation, transparency, and health-conscious decision-making. We have also plans to draw inspiration from recent advancements in generative recommender models
[
2
,
7
,
8
]

Read the complete text