Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Persona-aware and Explainable Bikeability Assessment: A Vision-Language Model Approach

Yilong Dai, Ziyi Wang, Chenguang Wang, Kexin Zhou, Yiheng Qian, Susu Xu, Xiang Yan

TL;DR

This work tackles perception-based bikeability assessment by introducing a persona-aware Vision-Language Model that conditions on cyclist typologies to generate explainable evaluations. It combines three data formats through multi-granularity instruction tuning and uses AI-enabled image augmentation to isolate the impact of infrastructure factors. A panoramic DC crowdsourcing dataset (12,400 persona-conditioned assessments from 427 cyclists) validates the approach, showing competitive rating predictions alongside unique factor attribution and factor identification capabilities. The framework advances human-centered transportation research by acknowledging rider heterogeneity and delivering interpretable outputs for planning decisions.

Abstract

Bikeability assessment is essential for advancing sustainable urban transportation and creating cyclist-friendly cities, and it requires incorporating users' perceptions of safety and comfort. Yet existing perception-based bikeability assessment approaches face key limitations in capturing the complexity of road environments and adequately accounting for heterogeneity in subjective user perceptions. This paper proposes a persona-aware Vision-Language Model framework for bikeability assessment with three novel contributions: (i) theory-grounded persona conditioning based on established cyclist typology that generates persona-specific explanations via chain-of-thought reasoning; (ii) multi-granularity supervised fine-tuning that combines scarce expert-annotated reasoning with abundant user ratings for joint prediction and explainable assessment; and (iii) AI-enabled data augmentation that creates controlled paired data to isolate infrastructure variable impacts. To test and validate this framework, we developed a panoramic image-based crowdsourcing system and collected 12,400 persona-conditioned assessments from 427 cyclists. Experiment results show that the proposed framework offers competitive bikeability rating prediction while uniquely enabling explainable factor attribution.

Persona-aware and Explainable Bikeability Assessment: A Vision-Language Model Approach

TL;DR

This work tackles perception-based bikeability assessment by introducing a persona-aware Vision-Language Model that conditions on cyclist typologies to generate explainable evaluations. It combines three data formats through multi-granularity instruction tuning and uses AI-enabled image augmentation to isolate the impact of infrastructure factors. A panoramic DC crowdsourcing dataset (12,400 persona-conditioned assessments from 427 cyclists) validates the approach, showing competitive rating predictions alongside unique factor attribution and factor identification capabilities. The framework advances human-centered transportation research by acknowledging rider heterogeneity and delivering interpretable outputs for planning decisions.

Abstract

Bikeability assessment is essential for advancing sustainable urban transportation and creating cyclist-friendly cities, and it requires incorporating users' perceptions of safety and comfort. Yet existing perception-based bikeability assessment approaches face key limitations in capturing the complexity of road environments and adequately accounting for heterogeneity in subjective user perceptions. This paper proposes a persona-aware Vision-Language Model framework for bikeability assessment with three novel contributions: (i) theory-grounded persona conditioning based on established cyclist typology that generates persona-specific explanations via chain-of-thought reasoning; (ii) multi-granularity supervised fine-tuning that combines scarce expert-annotated reasoning with abundant user ratings for joint prediction and explainable assessment; and (iii) AI-enabled data augmentation that creates controlled paired data to isolate infrastructure variable impacts. To test and validate this framework, we developed a panoramic image-based crowdsourcing system and collected 12,400 persona-conditioned assessments from 427 cyclists. Experiment results show that the proposed framework offers competitive bikeability rating prediction while uniquely enabling explainable factor attribution.
Paper Structure (31 sections, 7 equations, 6 figures, 3 tables)

This paper contains 31 sections, 7 equations, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Bikeability Assessment
  4. Vision-Language Models for Urban Analysis
  5. Persona-aware Modeling
  6. Method
  7. Data Collection and Preprocessing
  8. Study Design and Sampling
  9. Image Augmentation Pipeline
  10. Crowdsourcing survey
  11. Persona Classification
  12. Task Definition
  13. Multi-format Training Data
  14. Model Training
  15. Experiments
  16. ...and 16 more sections

Figures (6)

  • Figure 1: Overview of our persona-aware explainable bikeability assessment framework.
  • Figure 2: Three-stage model architecture: (1) Multi-source data collection with crowdsourced survey and AI-based image augmentation, (2) Multi-granularity supervised fine-tuning with three data types (Type 1: full reasoning, Type 2: factor-rating pairs, Type 3: rating-only), and (3) Preference-based reasoning refinement with DPO.
  • Figure 3: AI-based image augmentation: original street-view images (left) and systematically modified versions with controlled infrastructure changes (right).
  • Figure 4: Within-participant rating variance distribution across 427 participants. (a) Histogram shows substantial heterogeneity (median = 0.872, range: 0--2.17). (b) Scatter plot confirms variance is orthogonal to mean rating (r = $-$0.10).
  • Figure 5: Survey interface 1: immersive 360-degree Google Street View for bikeability assessment.
  • ...and 1 more figures