Photo2Trip: an interactive trip planning system based on geo-tagged photos

Huagang Yin, Xin Lu, Changhu Wang, Nenghai Yu, Lei ZhangACM Multimedia 2010 · DOI

TL;DR

This paper introduces Photo2Trip, an interactive trip planning system that leverages a massive dataset of geo-tagged photos to generate personalized travel itineraries. By analyzing millions of user-contributed photos and travelogues, Photo2Trip aims to overcome the limitations of traditional trip planning methods, offering a more efficient and engaging experience.

What Problem Does This Paper Address?

Planning a trip often involves sifting through vast amounts of unstructured information from travelogues, forums, and personal accounts. This process is time-consuming and can be inefficient, especially when trying to create a coherent and personalized itinerary that accounts for various preferences like duration, time of year, and travel style. While GPS data could offer precise travel paths, privacy concerns limit its widespread use for trip planning. The authors identify a gap in existing tools for automatically generating comprehensive and customizable travel routes that are both informative and engaging.

Methods

Photo2Trip employs a multi-stage approach:

  • Destination Discovering: Utilizes a MeanShift Clustering Algorithm on 20 million geo-tagged photos to identify over 300,000 potential destinations (attractions/landmarks). The top 10% largest clusters are retained. Destination styles and popular visiting times are inferred from 200,000 travelogues.
  • Internal Path Discovering: Addresses the sparsity of individual user photo footprints by merging photos from multiple tourists at the same location. This process identifies and ranks candidate paths within an attraction based on "path quality" and "path popularity," estimating stay times.
  • Trip Planning: Formulates trip planning as a graph analysis problem. Destinations become nodes, and transitions between them are edges. A dynamic programming algorithm finds optimal routes that maximize a score (likely related to popularity or user preference) within a given travel duration constraint.
  • User Interaction: The system offers an interactive interface allowing users to input preferences (location, duration, time, style) and then modify suggested plans by adding or removing attractions. The system dynamically regenerates routes based on these changes. Users can also explore internal paths within attractions, view representative photos, and see animations of the paths.

Datasets

  • Geo-tagged Photos: 20 million geo-tagged photos, primarily sourced from Panoramio (http://www.panoramio.com).
  • Travelogues: 200,000 travelogues crawled from weblogs and professional travel websites.

The paper does not explicitly mention newly introduced benchmarks, but the scale of the geo-tagged photo dataset and the resulting coverage of over 30,000 attractions in more than 100 countries can be considered a significant contribution.

Key Insights and Ideas

  • Leveraging Geo-tagged Photos for Trip Planning: The core insight is that publicly available geo-tagged photos, despite their inherent sparsity and incompleteness, contain rich enough information to reconstruct meaningful travel paths and plan itineraries.
  • Merging Incomplete Footprints: The technique of merging individual, incomplete photo footprints from multiple users is crucial for creating comprehensive internal paths within attractions, overcoming the limitations of single-user data.
  • Interactive Customization: The emphasis on user interaction, allowing users to easily modify and refine automatically generated plans, is key to achieving personalized travel experiences.
  • Multi-level Planning: The system supports planning at both the city level (order of attractions) and the attraction level (internal paths), providing a holistic approach to trip planning.

Key Results

  • The system supports trip planning for over 30,000 attractions/landmarks in more than 100 countries and territories.
  • Photo2Trip can automatically recommend popular travel routes and suggest typical internal paths within destinations.
  • Users can interactively adjust suggested plans, leading to customized travel routes.
  • The system provides representative photos, tags, and animations to help users understand attractions and paths.
  • Examples show generated routes for Washington D.C. and internal paths for the Lincoln Memorial and Forbidden City, demonstrating the system's capabilities.

Limitations and Future Work

  • Data Sparsity: While the system addresses data sparsity by merging photos, the inherent incompleteness of user-contributed data remains a challenge.
  • Privacy Concerns: Although the paper notes that GPS data is not used due to privacy, the reliance on publicly available photos means that the system's coverage might be biased towards popular tourist spots with more online photo sharing.
  • Algorithm Details: The paper refers to a specific algorithm [1] for recovering paths, but detailed explanations of this algorithm are not provided within this document.

Future work is not explicitly detailed, but the foundation laid by Photo2Trip suggests potential for further refinement of path reconstruction algorithms, integration of other data sources, and more sophisticated personalization based on user behavior.

Summary updated

BibTeX
@inproceedings{yin2010photo,
  author = {Huagang Yin and Xin Lu and Changhu Wang and Nenghai Yu and Lei Zhang},
  title = {{Photo2Trip: an interactive trip planning system based on geo-tagged photos}},
  booktitle = {ACM Multimedia},
  year = {2010},
  doi = {10.1145/1873951.1874291},
}
Figure