chuuhtetnaing/stravl-travel-preference-dataset

Source data from the Stravl/Stravl-Data GitHub repository. This dataset is a derived, HuggingFace-ready version with human-readable label columns added.

Stravl Travel Preference Dataset

Stravl-Data is one of the world's largest open-sourced datasets of travel preferences. Through a user-facing website, Stravl gathered 80,301 travelers' vacation preferences. Users were asked to fill out a brief form on their ideal vacation (such as expected experiences, scenery, and activity level) and logistical constraints (such as budget, traveler ages, and season). Thereafter, they "swiped" or "rated" ten destinations, responding with "Yes", "No", or "Maybe" in a Tinder-like rating framework. In total, over 1,000,000 user swipes were recorded. Lastly, users were shown 5 to 10 recommendations selected by different ML models; those recommendations and the users' feedback on them are stored.

Personally-identifying information such as IP data, user names, or other metadata has been removed prior to release.

What's different in this version

This version is derived from the original GitHub dataset for convenient use with the 🤗 datasets library. On top of the original columns, human-readable label columns have been added directly beside their coded counterparts:

• Each form column (form_a … form_rr) gets a descriptive sequence column resolving its codes to labels (e.g. form_b → form_b_trip_budget).
• Each swipe column (yes_swipes, no_swipes, maybe_swipes) gets a *_names column resolving destination indices to destination names.

The original coded columns are preserved unchanged.

Dataset Structure

Column	Type	Description
id	string	Anonymized user identifier
form_a … form_rr	string	Raw form responses, each a stringified list of codes (e.g. "['0', '1']")
form_a_age_ranges	sequence[string]	Decoded labels for form_a (travel group age ranges)
form_b_trip_budget	sequence[string]	Decoded labels for form_b (budget per person, per night)
form_c_travel_season	sequence[string]	Decoded labels for form_c (season)
form_f_experiences	sequence[string]	Decoded labels for form_f (desired experiences)
form_g_scenery	sequence[string]	Decoded labels for form_g (desired scenery)
form_h_activity_level	sequence[string]	Decoded labels for form_h (activity level)
form_i_safety_conscious	sequence[string]	Decoded labels for form_i (safety preference)
form_j_destination_popularity	sequence[string]	Decoded labels for form_j (destination popularity)
form_k	string	Raw form response (undocumented; preserved as-is, no decoded column)
form_r_destination_choice	sequence[string]	Decoded labels for form_r (anywhere vs. specific regions)
form_rr_specific_regions	sequence[string]	Decoded labels for form_rr (chosen regions)
yes_swipes / no_swipes / maybe_swipes	string	Stringified lists of destination indices the user swiped Yes / No / Maybe
yes_swipes_names / no_swipes_names / maybe_swipes_names	sequence[string]	Destination names corresponding to the swipe indices
Model / Retrieval / DynaMatch	int	Which recommendation algorithm(s) were used (-1 = not tracked)
Rating_0 … Rating_9	int	User feedback per recommendation: 1 = approved, -1 = disapproved, empty = not rated. All -1 = ratings not tracked for this user
Rec_0 … Rec_9	string	Recommended destination names, in recommendation order (-1 = not tracked)

Form response encodings

Notation — ordered scales read left→right (low → high); pick one is a plain list; pick many uses checkboxes; the code for each option is shown in code style.

Ordered scales

FORM_H — Activity level 0 Chill & relaxed → 1 Balanced → 2 Active

FORM_I — Safety conscious 0 Very conscious → 1 Balanced → 2 Ready for anything

FORM_J — Destination popularity 0 Off the beaten path → 1 Classic spot → 2 Mainstream & trendy

FORM_B — Trip budget (per person, per night) 0 $0-49 → 1 $50-99 → 2 $100-249 → 3 $300+

Budget has a gap: 2 ends at $249 and 3 starts at $300, so $250-299 maps to no code. The intended bucket was likely $250+.

Pick one

FORM_C — Travel season

• 0 Winter
• 1 Spring
• 2 Summer
• 3 Fall

FORM_R — Where do you want to go?

• 0 Anywhere
• 1 Specific regions → unlocks FORM_RR

FORM_RR — Which specific regions? (pick many · only if FORM_R = 1)

• e Europe
• n N. America
• c Caribbean
• a Asia
• s S. America
• m Mid. East
• f Africa
• o Oceania

Pick many

FORM_A — Age ranges in the travel group

• 0 0-19
• 1 20-39
• 2 40-59
• 3 60+

FORM_F — Experiences sought

• 0 Beach
• 1 Adventure
• 2 Nature
• 3 Culture
• 4 Nightlife
• 5 History
• 6 Shopping
• 7 Cuisine

FORM_G — Scenery sought

• 0 Urban
• 1 Rural
• 2 Sea
• 3 Mountain
• 4 Lake
• 5 Desert
• 6 Plains
• 7 Jungle

Undocumented

form_k appears in the source data but isn't documented. It's preserved unchanged — no decoded column is generated.

Swipe & Recommendation Responses

• Swipes: yes_swipes, no_swipes, and maybe_swipes are lists of indices into a destination table; the *_names columns resolve them to destination names.
• Recommendations & Ratings: Users were recommended 5 destinations (or 10 if they asked for more). The Model, Retrieval, and DynaMatch fields indicate which algorithms produced them; the algorithm implementations are not released. Rec_* holds the recommended destination names in order, and Rating_* holds user feedback on each (1 approve, -1 disapprove, empty = unrated; all -1 = ratings not tracked for that user).

Dataset Statistics

• Users: 80,301
• Total swipes: 1,000,000+
• Destinations: 1,486 unique destinations
• Language: English

Usage

from datasets import load_dataset

ds = load_dataset("chuuhtetnaing/stravl-travel-preference-dataset")

example = ds["train"][0]
print("Liked:", example["yes_swipes_names"])
print("Budget:", example["form_b_trip_budget"])
print("Experiences:", example["form_f_experiences"])

Limitations

• Form responses are self-reported and reflect each user's stated preferences at the time of the survey.
• Recommendation algorithm implementations (Model, Retrieval, DynaMatch) are not released — only their selections are shared.
• Recommendations and ratings were not tracked for all users (-1 markers indicate untracked entries).

License

This dataset is published under permissive terms inherited from the upstream source. Per the original Stravl documentation, the data is free to use for any academic or commercial purposes. There is no formal LICENSE file in the source repository; for questions about usage, contact info@stravl.com.

Acknowledgments

This dataset is derived from Stravl/Stravl-Data on GitHub. All credit for collecting and open-sourcing the original travel preference data goes to Stravl. This HuggingFace release adds human-readable label columns for convenience and is maintained as a fork of the original repository.