Resources by Stefan

Twitter US Airline Sentiment

Twitter US Airline Sentiment

Stefan
Open Dataset
Creators: (Makone, Ashutosh)
Publication Date: 2016
Creators: (Makone, Ashutosh)

The Twitter US Airline Sentiment dataset is a collection of tweets aimed at analyzing public sentiment toward major U.S. airlines. Compiled in February 2015, the dataset consists of 14,640 tweets directed at several U.S. airlines. It serves as a valuable resource for sentiment analysis and natural language processing research, particularly in understanding customer satisfaction, airline service quality, and issues reported by travelers. Each tweet in the dataset is labeled with one of three sentiment categories: positive, neutral, or negative. Tweets labeled as negative are further categorized into specific negative sentiment reasons, such as late flight, customer service issue, canceled flight, and lost luggage, providing deeper insights into common complaints. The dataset also identifies the airline mentioned in each tweet, covering six major U.S. carriers: United Airlines, US Airways, American Airlines, Southwest Airlines, Delta Air Lines, and Virgin America. Additional metadata is provided for each tweet, including tweet ID, tweet text, tweet coordinates (if available), user information, and location data, allowing for further contextual analysis. The dataset is relatively small, with a total size of 8,46 MB, making it easily manageable for sentiment analysis tasks and machine learning applications. It includes 14,640 tweets from 7,700 unique users, providing a broad yet concise representation of customer interactions with airlines on Twitter. The tweets were collected over a one-month period in February 2015, offering a snapshot of public sentiment during that specific timeframe.

MovieTweetings

MovieTweetings

Stefan
Open Dataset
Creators: Dooms, Simon; De Pessemier, Toon; Martens, Luc
Publication Date: 2013
Creators: Dooms, Simon; De Pessemier, Toon; Martens, Luc
MovieTweetings is a dataset consisting of ratings on movies that were contained in well-structured tweets on Twitter. The goal of this dataset is to provide the RecSys community with a live, natural and always up-to-date movie ratings dataset. The dataset has been actively collecting ratings since February 28, 2013, and will be updated as much as possible to incorporate rating data from the newest tweets available. The dataset includes 921,398 ratings from 71,707 unique users. The ratings contained in the tweets are scaled from 0 to 10, as is the norm on the IMDb platform. In total, the dataset has a size of 26,2 MB and consists of two main files:

  • ratings.dat contains extracted ratings, structured as:
    user_id::movie_id::rating::rating_timestamp

    • user_id: Unique identifier for the user.
    • movie_id: IMDb identifier for the movie.
    • rating: User’s score on a 10-star scale.
    • rating_timestamp: Unix timestamp when the rating was extracted.

  • items.dat includes metadata about the rated movies, structured as:
    movie_id::movie_title (movie_year)::genre|genre|genre

    • movie_id: IMDb identifier for the movie.
    • movie_title: Name of the movie along with the release year.
    • genre: Pipe-separated list of genres.
Stock return prediction with tweets

Stock return prediction with tweets

Stefan
Open Dataset
Creators: Madhyastha, Pranava; Sowinska, Karolina
Publication Date: 2020
Creators: Madhyastha, Pranava; Sowinska, Karolina

This dataset is designed to analyze the impact of Twitter-based textual information on stock returns. Compiled by researchers Karolina Sowinska and Pranava Madhyastha, this dataset was published in 2020 and is made available under the GNU General Public License v3.0 or later. It provides valuable data for financial analytics and natural language processing, particularly in studying the relationship between social media sentiment and stock market performance. By linking tweets to stock return data, the dataset enables the development of predictive models for stock movement based on public sentiment. The dataset comprises 862,231 labeled tweets, all in English, each associated with specific companies. These tweets serve as samples for analyzing public opinion and sentiment regarding different stocks and financial events. A cleaned subset of 85,176 labeled instances is also included, making the dataset suitable for both large-scale machine learning models and more focused analyses. Each tweet is linked to corresponding stock return data, allowing for a company-level examination of how Twitter sentiment impacts one-day, two-day, three-day, and seven-day stock returns. This structured linkage between tweets and financial performance provides a unique opportunity to study the effects of social media on stock price fluctuations. The dataset is approximately 225 MB in size on GitHub, making it manageable for various analytical tasks, including sentiment analysis, text-based predictive modeling, and financial forecasting. It is structured into two primary components:

  • Tweet Data: This includes the textual content of tweets, user metadata, timestamps, and the companies referenced in each tweet. These features allow researchers to perform sentiment analysis, track user engagement, and examine the frequency of stock-related discussions on social media.

  • Stock Return Data: This includes numerical stock return values corresponding to the companies mentioned in the tweets. The returns are recorded over multiple time intervals, enabling the study of both short-term and long-term price movements in response to social media discussions.

IMDb movies extensive dataset

IMDb movies extensive dataset

Stefan
Open Dataset
Creators: (Leone, Stefano)
Publication Date: 2019
Creators: (Leone, Stefano)

The movies dataset serves as a valuable resource for researchers, data analysts, and movie enthusiasts looking to explore various aspects of cinema. It contains detailed metadata on movies, including ratings, cast and crew details, and audience reception, making it highly suitable for studies related to film trends, genre popularity, audience preferences, and predictive modeling of movie success. The dataset covers movies up to the year 2019, providing a broad temporal range that allows for longitudinal studies on film industry trends and developments. In total, it has a size of approximately 230 MB and includes 85,855 movies with attributes such as movie description, average rating, number of votes, genre, etc. The dataset is structured into multiple components, each offering specific insights: The ratings dataset includes 85,855 rating details from demographic perspective. The names dataset includes 297,705 cast members with personal attributes such as birth details, death details, height, spouses, children, etc. The title principals dataset includes 835,513 cast members roles in movies with attributes such as IMDb title id, IMDb name id, order of importance in the movie, role, and characters played. By offering a rich and detailed collection of movie-related information, the IMDb Movies Extensive Dataset is useful for researchers, film industry professionals, and data scientists looking to gain deeper insights into the world of cinema.

Twitter Dataset

Twitter Dataset

Stefan
Open Dataset
Creators: Cheng, Zhiyuan; Caverlee, James; Lee, Kyumin
Publication Date: 2010
Creators: Cheng, Zhiyuan; Caverlee, James; Lee, Kyumin
This dataset is a collection of scraped public twitter updates used in coordination with an academic project to study the geolocation data related to twittering. We provide both training set and test set in the paper You Are Where You Tweet: A Content-Based Approach to Geo-locating Twitter Users in CIKM 2010. The training set contains 115,886 Twitter users and 3,844,612 updates from the users. All the locations of the users are self-labeled in United States in city-level granularity. The test set contains 5,136 Twitter users and 5,156,047 tweets from the users. In total, the dataset has a size of 30,0 kB. All the locations of users are uploaded from their smart phones with the form of “UT: Latitude,Longitude”. The Twitter activity is covered over a period of five months, from September 2009 to January 2010, offering a valuable temporal snapshot of user interactions and content generation during that time. Structurally, the dataset is divided into four text files. The training set users file (“training_set_users.txt”) contains user information in the format “UserIDtUserLocation”, and the training set tweets file (“training_set_tweets.txt”) stores tweets in the format “UserIDtTweetIDtTweettCreatedAt”. Similarly, the test set users file (“test_set_users.txt”) follows the same format as the training set users file, while the test set tweets file (“test_set_tweets.txt”) follows the same structure as the training set tweets file.
Customer Support on Twitter

Customer Support on Twitter

Stefan
Open Dataset
Creators: Axelbrooke, Stuart
Publication Date: 2017
Creators: Axelbrooke, Stuart

The Customer Support on Twitter dataset is a large, modern corpus of tweets and replies to aid innovation in natural language understanding and conversational models, and for study of modern customer support practices and impact. It is intended to facilitate advancements in natural language understanding and the development of conversational models. Compiled by Stuart Axelbrooke in 2017, this dataset encompasses tweets and replies from prominent companies such as Apple, Amazon, Uber, Delta, and Spotify. It provides valuable insights into contemporary customer support practices and their impact, making it an essential resource for researchers interested in automated response generation, sentiment analysis, and conversational flow modeling. The dataset is approximately 516.53 MB in size. It is designed for the analysis of conversation dynamics and contains several key attributes. Each tweet entry has a unique, anonymized tweet ID (tweet_id), an anonymized user ID (author_id), a timestamp (created_at), and the tweet text (text), where sensitive information such as phone numbers and email addresses has been masked to ensure privacy. It differentiates between inbound tweets (inbound), which are directed at companies by customers, and outbound tweets, which are responses from the companies. Additionally, in_response_to_tweet_id and response_tweet_id fields allow for the reconstruction of entire conversation threads by linking tweets to their respective responses.

Spotify Million Playlist Dataset

Spotify Million Playlist Dataset

Stefan
Open Dataset
Creators: Chen, Ching-Wei; Lamere, Paul ; Schedl, Markus ; Zamani, Hamed
Publication Date: 2018
Creators: Chen, Ching-Wei; Lamere, Paul ; Schedl, Markus ; Zamani, Hamed
We released a dataset of one million user-created playlists from the Spotify platform, dubbed the Million Playlist Dataset (MPD). The dataset includes, for each playlist, its title as well as the list of tracks (including album and artist names), and some additional metadata such as Spotify URIs and the playlist’s number of followers. The dataset has a size of 5,39 GB and contains 1,000,000 playlists, including playlist titles and track titles, created by users on the Spotify platform between January 2010 and October 2017. It is ideal for building and evaluating recommendation algorithms, studying user behavior in music consumption, and understanding how playlists evolve over time. The dataset is widely used by researchers and developers to improve machine learning models for music streaming applications, ensuring a more personalized and engaging experience for users.
The Music Streaming Sessions Dataset

The Music Streaming Sessions Dataset

Stefan
Open Dataset
Creators: Brost, Brian; Mehrotra, Rishabh; Jehan, Tristan
Publication Date: 2018
Creators: Brost, Brian; Mehrotra, Rishabh; Jehan, Tristan

The MSSD is a large-scale collection of user interaction data from a music streaming service, designed to support research in user behavior modeling, music information retrieval, and session-based recommendation systems. Released in 2019, this dataset contains approximately 160 million listening sessions, making it one of the most extensive datasets available for analyzing how users engage with music streaming platforms. It provides valuable insights into listening habits, session structures, and sequential user interactions, enabling researchers to study music recommendation, user retention, and engagement patterns. The dataset has a size of 70 GB and captures approximately 3.7 million unique tracks, covering a diverse range of musical content. Each session includes detailed user interactions, such as play, pause, skip, and seek actions, offering a granular view of how listeners interact with music over time. Additionally, it contains rich metadata and audio features for each track, including details such as track ID, artist name, album name, and genre, along with audio attributes like tempo, key, and loudness. These elements make the dataset highly valuable for both behavioral studies and technical research in music information retrieval.

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.