Source-linked AI summary
The Shortest Path to Happiness: Recommending Beautiful, Quiet, and Happy Routes in the City
Daniele Quercia, Rossano Schifanella, Luca Maria Aiello
TL;DR
The paper addresses the gap between shortest-path routing and pedestrians’ preferences for emotionally pleasant urban routes. It combines crowdsourced perceptions with graph-based path search, finding that recommended routes improve emotional qualities while adding only a few walking minutes, and that Flickr metadata can provide useful beauty proxies in London and Boston.
Problem
Mapping services recommend shortest routes, while prior route research had not considered people’s emotional perceptions of urban spaces when recommending routes.
Method
The paper builds a location graph, converts crowdsourced beauty, quietness, and happiness judgments into scores, and selects paths balancing shortness with pleasantness.
Results
Recommended routes were emotionally pleasing and added only a few extra walking minutes compared with shortest routes; Flickr metadata also provided beauty proxies tested in London and Boston.
Takeaways & Limitations
Emotion-aware routing can recommend more pleasant urban paths without requiring substantially longer walks, and new-city recommendations need not always begin with crowdsourced ratings.
Takeaways & Limitations
Perceptions of urban qualities differ between individuals, and cell-based spatial modeling cannot capture conditions inside cells or the links between cell centroids.
Abstract
from arXiv · showhide
When providing directions to a place, web and mobile mapping services are all able to suggest the shortest route. The goal of this work is to automatically suggest routes that are not only short but also emotionally pleasant. To quantify the extent to which urban locations are pleasant, we use data from a crowd-sourcing platform that shows two street scenes in London (out of hundreds), and a user votes on which one looks more beautiful, quiet, and happy. We consider votes from more than 3.3K individuals and translate them into quantitative measures of location perceptions. We arrange those locations into a graph upon which we learn pleasant routes. Based on a quantitative validation, we find that, compared to the shortest routes, the recommended ones add just a few extra walking minutes and are indeed perceived to be more beautiful, quiet, and happy. To test the generality of our approach, we consider Flickr metadata of more than 3.7M pictures in London and 1.3M in Boston, compute proxies for the crowdsourced beauty dimension (the one for which we have collected the most votes), and evaluate those proxies with 30 participants in London and 54 in Boston. These participants have not only rated our recommendations but have also carefully motivated their choices, providing insights for future work.
1. INTRODUCTION
The paper argues that route recommenders should account for emotional qualities, not only travel time. It proposes crowdsourced urban-perception scores and evaluates routes optimized for beauty, quietness, and happiness.
- Motivation: Mapping services typically recommend shortest routes, even when pedestrians prefer slower alternatives with pleasant scenery.The motivation contrasts fast routing with preferences for tree-lined small streets over car-dominated avenues.
- Research gap: Prior route research has not incorporated people’s emotional perceptions of urban spaces into recommendations.The paper connects this gap to psychogeography’s focus on how geographical environments affect emotions and behavior.
- Approach: The authors build a location graph and rank places using crowdsourced judgments of beauty, quietness, and happiness.Users compare pairs of London street scenes and vote on which scene better expresses each quality.
- Evaluation: The proposal is quantitatively validated for recommending routes that are short and emotionally pleasing, then qualitatively evaluated with 30 London participants.The evaluation examines whether recommendations achieve the intended emotional properties and how users assess them.
- Generalizability: The study tests generalizability by predicting beauty from Flickr metadata and evaluating beauty-derived routes with participants in London and Boston.The Boston study uses a new group of 54 participants.
2. RELATED WORK
Earlier route-recommendation work emphasized efficiency, distinctive urban experiences, or personalization. This paper identifies emotional perception as an unaddressed basis for recommending urban routes.
- Efficiency-oriented routing: Early systems optimized efficient car and public-transport routes, while later systems addressed safety and cycling smoothness.Examples include backtracking, adaptive A*-like routing, and recommendations for safer or smoother cycle paths.
- Interesting routes: Other work used geo-referenced online content, especially Flickr, to learn popular or distinctive urban trajectories.These approaches mined spatiotemporal sequences of picture uploads to recommend corresponding routes.
- Location discovery: GPS traces can identify interesting places beyond cities, but prior work focused on locations rather than routes.One approach ranked places through graph centrality in a bipartite graph of users and visited locations.
- Tourism scope: Touristic route methods tolerate substantially longer paths because frequently visited attractions are reasonable recommendations in those contexts.The paper distinguishes these use cases from ordinary urban walking.
- Personalization: Personalization research incorporated location demands or demographic information into route planning for tourists and residents.Examples include itineraries constrained by location types and Bayesian models using demographic annotations.
- Research gap: No prior work had considered emotional perceptions of urban spaces when recommending routes, motivating this paper’s collection, modeling, and evaluation of such perceptions.The authors combine perception collection with algorithmic routing and quantitative and qualitative evaluation.
3. OUR PROPOSAL
The proposal represents a city as a location graph, converts crowdsourced emotional judgments into location scores, and searches shortest-path alternatives for routes balancing pleasantness and length.
- 3. OUR PROPOSAL: The method seeks a short, pleasant path from source s to destination d through graph construction, perception collection, scoring, and path selection.The final path balances travel distance against emotional pleasantness.
- 3.1 Building Location Graph: Central London is divided into 532 walkable cells of 200x200 meters, with each cell represented as a graph node linked to eight geographic neighbors.The 200-meter unit corresponds approximately to two blocks or a 2.5-minute walk in dense London.
- 3.2 Crowdsourcing Perceptions: UrbanGems compares pairs of Google Street View and Geograph scenes, asking users which is more beautiful, quiet, or happy.The design uses comparable-quality images rather than Flickr images and includes controls for image bias and answer sparsity.
- 3.2 Crowdsourcing Perceptions: 3,301 participants supplied 17,261 annotation rounds used to rank pictures along beauty, quietness, and happiness.Participants came from London, the rest of the UK, and outside the UK; beauty received preferentially more answers because it was the default question.
- 3.2 Crowdsourcing Perceptions: Beauty and happiness scores correlate most strongly, with beauty-happy r = 0.64, compared with quiet-beauty r = 0.33 and happy-quiet r = 0.29.All reported correlations are statistically significant with p-values < 0.0001.
- 3.3 Scoring Locations: A location’s pleasantness likelihood is modeled as proportional to the probability that visiting it produces the relevant emotional perception.Crowdsourced scores are transformed into probabilities using functions such as a cubic curve, then applied to happiness, quietness, and beauty.
- 3.4 Selecting Best Path: The path search generates M shortest alternatives with Eppstein’s algorithm, iteratively evaluates the first m paths, and retains the path with the best average emotional rank.The procedure stops when marginal rank improvement falls below a threshold selected using the Marginal Value Theorem, producing beauty, quiet, happiness, and shortest baseline paths.
- 3.4 Selecting Best Path: Figure 2 compares emotional-rank improvement with extra walking length as the exploration level m increases.The left panel shows beauty, quiet, and happiness rank gains; the right panel shows added walking time when considering the first m shortest paths.
4. EVALUATION
The evaluation tests whether emotionally pleasant routes improve beauty, quietness, and happiness without substantially increasing distance. Quantitative comparisons and a mixed-method user study assess route quality, length trade-offs, and perceived desirability.
- Validation: 30% more beautiful, 26% quieter, and 30% happier were achieved by corresponding recommended paths compared with shortest paths.Results average 190 paths between 20 landmark nodes.
- Length Trade-off: 12% longer on average, or roughly 7.5 additional walking minutes, were the recommended paths compared with shortest paths.The average combines both long and short paths.
- Length Trade-off: 40% longer paths were typical for 1 km destinations, while the extra cost decreased exponentially as destination distance increased.Shorter routes require more deviations to locate pleasant scenery; longer routes provide more room to include it.
- User Assessment: Four path variations—shortest, beautiful, quiet, and happy—were compared across three scenarios with fixed endpoints between Euston Square and Tate Modern.Fixed endpoints were used to avoid rating sparsity, and the locations were chosen for familiarity, walkability, and the Thames crossing.
- User Assessment: The shortest path performed worst across scenarios, with median ratings from 1.5 to 2 and dispersion of ±1.Participants evaluated the routes using Likert responses in beauty, quiet, and happiness scenarios.
- User Assessment: A quiet path received 3±0 versus 1±1 for the shortest path, while the happy variation reached a median of 3.Participants associated routes with their intended qualities without being told which path represented which quality.
- User Assessment: Participants linked beautiful and happy paths with peaceful or historical places, while traffic-heavy or crowded streets made routes unpleasant.Frequently mentioned examples included back streets near St Paul’s, the Millenium Bridge, Kingsway, Fleet Street, and Bloomsbury.
5. CROWDSOURCING AND LONDON
The paper extends emotionally pleasant routing beyond crowdsourced ratings by predicting beauty from Flickr metadata, then tests the resulting routes in London and Boston. Flickr-derived routes improve beauty relative to shortest paths, although London respondents evaluate the recommendation neutrally while Boston respondents rate it positively.
- Beyond Crowdsourcing: Flickr: Two main drawbacks motivate the extension: crowdsourced ratings require intense user involvement, and the evaluation had focused on a single city.The paper addresses these concerns partly by predicting beauty from Flickr metadata and testing in both London and Boston.
- Beyond Crowdsourcing: Flickr: 3.7M Flickr pictures with classified tags support location-level features including picture density, views, favorites, comments, tags, and LIWC category counts.The study starts from a random sample of 7M London pictures; 3.7M remain usable after tag classification.
- Beyond Crowdsourcing: Flickr: beauty′ = 0.37 + 0.03 · log(density) + 0.20 · fp −0.21 · fn predicts beauty from picture density and positive versus negative-emotion tag fractions.The negative-emotion fraction includes ‘negemo’, ‘swear’, ‘anx’, ‘sad’, and ‘anger’.
- Beyond Crowdsourcing: Flickr: 31% of beauty-score variability is explained by Flickr tags using the linear regression.The authors choose linear regression as a preliminary, explainable Flickr model and acknowledge that more sophisticated models could exploit spatial autocorrelation.
- Results: 28% more beautiful than shortest paths is the average improvement for Flickr-derived London routes.In the London user study, 30 respondents rated the Flickr-derived route neutrally, with median 2 and 0 dispersion.
6. DISCUSSION
The discussion identifies scalability, personalization, spatial representation, and picture salience as practical considerations, while proposing richer applications that incorporate points of interest and personal memories.
- Scalability: The route-finding step is computationally efficient, with graph construction performed once per city and path computation handled by Eppstein’s algorithm.The authors report that a 24-core server computes the best path in 51 milliseconds.
- Personalization: Personalized recommendations remain necessary because perceptions of happiness and other urban qualities differ between individuals.Respondents disagreed about whether busy shopping streets were pleasant, motivating tailoring routes to users’ past visits.
- Limited Spatial Representation: The cell-based graph cannot represent conditions within cells or the character of links between neighboring centroids.The authors suggest more effective spatial representations, including Space Syntax, to capture omitted aspects such as walkability.
- Salient City Pictures: Flickr-based predictive models may use pictures that are not salient representations of their geographically tagged locations.The authors suggest computer-vision methods to estimate how well each picture represents its associated urban setting.
- Limited Contextual Representation: Future work could model how weather and time change the perceived pleasantness of places.Participants noted that the same area could feel lovely when quiet but manic at busier times.
- Beyond route recommendations: Applications could enrich routes with points of interest, historical imagery, and personal memories associated with places along the path.The discussion links these features to respondents’ interest in history, events, people, and personal stories.
7. CONCLUSION
The paper proposes emotionally pleasing urban routes and validates that they achieve their intended qualities with only a small increase over shortest routes. It also shows that Flickr metadata can provide workable beauty proxies for new cities, while noting that viewing a route does not capture the full experience of being there.
- Conclusion: The proposed route variations recommend emotionally pleasing paths while adding only a few extra walking minutes compared with shortest routes.Validation tested whether routes targeting qualities such as perceived happiness achieved their intended properties.
- Conclusion: 30 London participants perceived the recommendations as expected, and their quantitative and qualitative assessments confirmed the validation results.The study used both numerical ratings and participants’ motivated feedback.
- Conclusion: Flickr metadata can provide reasonable beauty proxies without collecting crowd-sourced ratings for every new city.The approach was tested in Boston with 54 participants.
- Conclusion: Viewing a path does not capture the full affective experience of being there.The authors plan future in-the-wild testing with a mobile application across cities in Europe and the USA.