SPaFE: A Crowdsourcing and Multimodal Recommender System to Ensure Travel Safety in a City

Abstract

Existing navigation applications such as Google Maps, Apple maps provide a core service to users to find the shortest path or the path that takes the least amount of time towards a user's destination. The applications and other research efforts have been sought to include other features such as 3D maps, neighboring facilities, traffic information, and multi-modal alternate route suggestion based on user constraints. None of these, however, take into account an important factor: "user safety while commuting". Although the common perception of street crime is that it is primarily a problem in the third-world countries, current popular hashtags or topics (e.g. "blacklifematters") indicate that it is now prevalent in other parts of the world as well. Even existing multi-modal alternative route recommender systems are incapable of adapting to a dynamic set of safety features, are unable to provide new safe path updates in response to real-time commuter responses, and take little or no account of historical on-road events when designing the safe algorithm. In light of the above background, we begin by developing a generic framework, namely On-road Risk Minimization Problem (ORMP). We then introduce a dynamic population-based algorithm, that we call Safe Path for Everyone (SPaFE), that solves ORMP using multi-modal historical and crowdsourced data. Finally, our extensive empirical results demonstrate that SPaFE markedly outperforms the state-of-the-art.

Elevating AI Research at Cognitive Agents and Interaction Lab: A Technical Report on High-Performance Computing Infrastructure

“…With a star�ng focus on Distributed Constraint Op�miza�on Problems [1] [2] [3], CAIL has grown to be recognized for its role in advancing distributed problem-solving for mul�-agent systems [4]. As the interests of our researchers have evolved, our team later started exploring other compelling areas, including Mul�-Agent Path Finding [5], mul�-modal recommender systems [6] [7], causal inference [8], deep reinforcement learning [9] [10], Graph Neural Networks [11], and others [12] [13].…”

CrowdSPaFE: A Crowd-Sourced Multimodal Recommendation System for Urban Route Safety

“…None of the above algorithms take into account multiple conflicting safety factors, have no user data, and do not train on local historical data. Motivated by these limitations "SPaFE" [18] introduced a population-based algorithm based on multi-modal local historical and crowd-sourced data. It codifies multiple conflicting safety factors to construct a single safety value.…”