Resources by Stefan

Google Restaurants

Google Restaurants

Stefan
Open Dataset
Creators: Zhankui He, Yan; Li, Jiacheng; Zhang, Tianyang; McAuley, Julian
Publication Date: 2022
Creators: Zhankui He, Yan; Li, Jiacheng; Zhang, Tianyang; McAuley, Julian

This is a mutli-modal dataset of restaurants from Google Local (Google Maps). Data includes images and reviews posted by users, as well as other metadata for each restaurant. The rich combination of textual reviews, numerical ratings, and visual content helps to provide a holistic view of user experiences and restaurant characteristics. Such a multi-faceted dataset is particularly valuable for developing and testing recommendation systems, conducting sentiment analysis, and exploring the relationships between visual content and user perceptions in the context of dining establishment. The total size of the dataset is approximately 120 GB and structured into:​

  • Restaurant Metadata: Information such as restaurant names, locations, contact details, and operational hours.​

  • User Reviews: Textual feedback and numerical ratings provided by users.​

  • Images: Photographs uploaded by users, showcasing various aspects of the restaurants.

Behance Community Art Data

Behance Community Art Data

Stefan
Open Dataset
Creators: He, Ruining; Fang, Chen; Wang, Zhaowen; McAuley, Julian
Publication Date: 2016
Creators: He, Ruining; Fang, Chen; Wang, Zhaowen; McAuley, Julian

Being a small, anonymized, version of a larger proprietary dataset, this dataset covers likes and image data from the community art website Behance. It provides valuable insights into user engagement with digital art, making it a significant resource for research in recommender systems, social network analysis, and the study of artistic preferences. Also, the dataset captures user interactions in the form of “appreciations” (akin to likes) on various art items. Each appreciation reflects a user’s positive acknowledgment of an artwork, offering a measurable indicator of engagement. Additionally, the dataset includes image features extracted from the artworks, facilitating analyses that combine user behavior with visual content characteristics.

In total, the dataset is about 3.5 GB large and encompasses:

  • Users: 63,497
  • Items: 178,788
  • Appreciates (“likes”): 1,000,000

The dataset is structured to include:​

  • User Data: Anonymized identifiers representing individual users.​

  • Item Data: Identifiers for each artwork, accompanied by associated image features.​

  • Appreciation Data: Records of user-item interactions, indicating which user appreciated which artwork.

Pinterest Fashion Compatibility

Pinterest Fashion Compatibility

Stefan
Open Dataset
Creators: Kang, Wang-Cheng; Kim, Eric; Leskovec, Jure; Rosenberg, Charles; McAuley, Julian
Publication Date: 2019
Creators: Kang, Wang-Cheng; Kim, Eric; Leskovec, Jure; Rosenberg, Charles; McAuley, Julian

This dataset is a structured collection of images and metadata designed to study the compatibility of fashion products within real-world scenes. It enables detailed analysis of how fashion items appear in different settings and supports applications in machine learning, recommendation systems, and virtual styling tools. One of its key features is the scene-product pairing, where fashion items in real-world images are annotated with bounding boxes and linked to corresponding product images. In total, the dataset includes 47,739 scene images, 38,111 product images, and 93,274 scene-product pairs, making it a comprehensive resource for fashion compatibility research.

The dataset is about 29 MB large and includes:

  • Scenes: 47,739
  • Products: 38,111
  • Scene-Product Pairs: 93,274
EndoMondo Fitness Tracking Data

EndoMondo Fitness Tracking Data

Stefan
Open Dataset
Creators: Ni, Jianmo; Muhlstein, Larry; McAuley, Julian
Publication Date: 2019
Creators: Ni, Jianmo; Muhlstein, Larry; McAuley, Julian

This is a collection of workout logs from users of EndoMondo. It contains sequential sensor data such as GPS coordinates (latitude, longitude, altitude), heart rate measurements, speed, and distance, making it valuable for studying workout patterns, performance tracking, and personalized fitness recommendations. Additionally, it includes user metadata such as anonymized user IDs, gender, and sport type, along with contextual factors like weather conditions. The dataset has a size of approximately 2.9 GB and consists of 1,104 users with 253,020 recorded workouts.

The dataset covers multiple components:​

  • User Information: Anonymized user identifiers and gender.​

  • Workout Details: Each workout log includes sport type, sequential data for GPS coordinates (latitude, longitude, altitude) with timestamps, heart rate measurements, and derived metrics such as speed and distance.

CrowdTangle Platform and API

CrowdTangle Platform and API

Stefan
Open Dataset
Creators: Garmur, Matt; King, Gary; Mukerjee, Zagreb; Persily, Nate; Silverman, Brandon
Publication Date: 2019
Creators: Garmur, Matt; King, Gary; Mukerjee, Zagreb; Persily, Nate; Silverman, Brandon

This document describes the CrowdTangle API and user interface being provided to researchers
by Social Science One under its collaboration framework with Facebook. CrowdTangle is a
content discovery and analytics platform designed to give content creators the data and insights
they need to succeed. This dataset enables users to monitor public content interactions, track trends, and identify influential accounts. The CrowdTangle API surfaces stories, and data to measure their social performance and identify influencers. This codebook describes the data’s scope, structure, and fields.

CrowdTangle’s dataset offers insights into public posts made by pages, groups, or verified profiles that have either surpassed 100,000 likes since 2014 or have been tracked by any active API user. The dataset includes all public posts from pages, groups, or verified profiles meeting the aforementioned criteria since 2014.

Key features include:

  • Content Discovery: Access to real-time data on trending posts, facilitating the identification of viral content and emerging topics.​

  • Performance Analytics: Metrics such as likes, shares, comments, and interaction rates, allowing for the assessment of content engagement.​

  • Influencer Identification: Tools to pinpoint accounts with significant influence within specific niches or broader audiences.

Facebook URL Shares

Facebook URL Shares

Stefan
Open Dataset
Creators: Solomon Messing; Bogdan State; Chaya Nayak; Gary King; Nate Persily
Publication Date: 2018
Creators: Solomon Messing; Bogdan State; Chaya Nayak; Gary King; Nate Persily

The data describes web page addresses (URLs) that have been shared on Facebook starting January 1, 2017 and ending about a month before the present day. URLs are included if shared by at least 20 unique accounts, and shared publicly at least once. We estimate the full data set will contain on the order of 2 million unique urls shared in 300 million posts, per week. By doing so, this dataset provides insights into the dissemination of web content on Facebook, capturing the dynamics of how information spreads across the platform. Researchers can use this data to explore patterns in user engagement, the virality of content, and the reach of various web pages within the Facebook ecosystem. The dataset’s focus on URLs shared by a minimum number of unique accounts ensures that the data represents content with a certain level of engagement, filtering out less significant shares.

The dataset is structured to include the following key components:​

  • URL Information: Each entry includes the web page address (URL) that was shared on Facebook.

  • Share Metrics: Data on the number of times each URL was shared, including the count of unique accounts that shared it and the total number of posts containing the URL.​

  • Engagement Metrics: Information on user interactions with the shared URLs, such as likes, comments, and shares.

Facebook Ad Library

Facebook Ad Library

Stefan
Open Dataset
Creators: Franklin Fowler, Erika; Franz, Mike; King, Gary; Martin, Greg; Mukerjee, Zagreb; Persily, Nate
Publication Date: 2019
Creators: Franklin Fowler, Erika; Franz, Mike; King, Gary; Martin, Greg; Mukerjee, Zagreb; Persily, Nate

The Ad Library API provides programmatic access to the Facebook Ad Library, a collection of all political advertisements run on Facebook and Instagram since May 2018 in the US, and for other dates in different countries. The codebook describes the scope, structure, and fields of these data. The Ad Library offers detailed information about each advertisement, including:​

  • Ad Creative: Visual and textual content of the ad.​

  • Impressions: Number of times the ad was displayed.​

  • Spend: Estimated amount spent on the ad.​

  • Demographics: Age, gender, and location breakdown of the audience reached.

Given that the Ad Library archives all ads related to political content, social issues, and elections since May 2018, the number of observations runs into the millions. The Ad Library’s data is structured to include various attributes for each advertisement:​

  • Ad ID: Unique identifier for each ad.​

  • Page ID and Name: Information about the page running the ad.​

  • Ad Creative: Content and format of the ad.​

  • Impressions and Spend: Metrics indicating the ad’s reach and budget.​

  • Demographic Distribution: Breakdown of the audience by age, gender, and location.

​

Multi-aspect Reviews

Multi-aspect Reviews

Stefan
Open Dataset
Creators: Julian McAuley; Jure Leskovec; Dan Jurafsky
Publication Date: 2013
Creators: Julian McAuley; Jure Leskovec; Dan Jurafsky
These datasets include reviews with multiple rated dimensions.It is particularly valuable for research in sentiment analysis, recommender systems, and user modeling, as it allows for a nuanced understanding of user opinions beyond overall ratings.​The most comprehensive of these are beer review datasets from Ratebeer and Beeradvocate, which include sensory aspects such as taste, look, feel, and smell. The data set is about 1 GB large.
Ratebeer:

  • Number of users: 40,213
  • Number of items: 110,419
  • Number of ratings/reviews: 2,855,232
  • Timespan: April, 2000 – November, 2011

BeerAdvocate:

  • Number of users: 33,387
  • Number of items: 66,051
  • Number of ratings/reviews: 1,586,259
  • Timespan: January, 1998 – November, 2011

The datasets are structured in a JSON format, with each entry representing a single review that includes:​

  • Product Information: Details about the beer being reviewed.​

  • User Information: Anonymized identifiers of the reviewers.​

  • Review Content: Textual feedback provided by the user.​

  • Ratings: Numerical scores for overall satisfaction and specific aspects (appearance, aroma, palate, taste).

Sign In

New to BERD@NFDI? Register

Forgot password?

Register

Reset Password

Please enter your username or email address, you will receive a link to create a new password via email.