ICWSM Data Challenge

Overview

The ICWSM Data Challenge is part of the ICWSM Data Sharing Initiative associated with the AAAI ICWSM conference. It provides large-scale, openly available social media and web datasets used in ICWSM-published research for use in annual research competitions and broader community work.

Key Details

• Type: Research data challenge / dataset sharing service
• Organizer: AAAI International Conference on Web and Social Media (ICWSM)
• Focus: Social media and web data (e.g., Twitter, blogs, social platforms)
• Access: Requires a registration process to obtain datasets
• Licensing: Datasets are released as community resources (openly available, subject to registration and any dataset-specific terms)

Features

• Dataset Hosting Service

  • Central hosting of datasets introduced by papers published in ICWSM proceedings.
  • Aims to make research datasets reusable by the wider community.

• Open Community Resources

  • All hosted datasets are released as community resources, intended for open research use.
  • Access governed via a registration process (details provided on the site under “obtaining”).

• Continuously Growing Collection

  • Initiative active for multiple years (e.g., ICWSM-16 noted as the fifth year).
  • Datasets from multiple conference editions (e.g., 2012, 2015, 2016 and beyond) are or will be made available.

• Social Media–Focused Datasets

  • Emphasis on large-scale collections of tweets and other social/web content.
  • Examples of 2015-related datasets (described at a high level by associated papers):
    • Twitter Streaming API vs. Firehose comparison.
    • Dynamics of emergent hashtags (#Bigbirds).
    • Detecting comments on news articles in microblogs.
    • Citation cascades in the blogosphere.
    • User-generated comments for social media object annotation.
    • Multi-indicator tweet geolocation.
    • Meme competition and success (Quickmeme.com).
    • Artist popularity across web and social music services.
    • Diurnal activity patterns from social media.
    • Political orientation inference from Twitter.
    • Political leaning quantification from tweets and retweets.

• Structured Dataset Descriptions (Example: ICWSM 2012)

  • Some years provide tabular metadata per dataset, including:
    • Number of files.
    • Number of observations (tweets, accounts, entries, etc.).
    • Number of Twitter users.
    • Network properties where applicable (nodes, edges).
  • Example 2012 datasets include (titles from associated papers):
    • Opinion retrieval in Twitter (tweets labeled as relevant/irrelevant for 50 queries).
    • Target-dependent sentiment expressions for movies and persons.
    • Home location inference from geo-tagged tweets in 100 top cities.
    • Conversation practices and network structure around a TV show (#XFactor Italia).
    • Impact and influence of bots on a social network (anobii.com social data).
    • Managing bad news in social media (Domino’s Pizza crisis tweet collection).
    • Tracking sentiment and topic dynamics from social media (Mozilla add-on reviews; description truncated in source but indicates review data).

• Conference Integration

  • Datasets are directly tied to specific ICWSM papers and years, facilitating reproducibility and follow-up work.
  • Supports annual research competitions and challenges run in conjunction with the conference.

Access and Registration

• Access requires following a registration process described on the ICWSM dataset page (section “obtaining”).
• Once registered, users can download available datasets from specific ICWSM years (e.g., 2012, 2015, 2016, etc.).

Use Cases

• Reproducibility of published ICWSM research.
• Benchmarking new algorithms on established social media datasets.
• Comparative studies across different social media phenomena (hashtags, political communication, sentiment, geolocation, network analysis, etc.).

Pricing

• The provided content only indicates that datasets are released as openly available community resources.
• No pricing or paid plans are mentioned; access appears to be free, subject to registration and any dataset-specific conditions.