Ò»±¾µÀÎÞÂë

INFORMATION ON TOOLS

Contact: yangkc@iu.edu

Description: Botometer checks the activity of a Twitter account and gives it a score. Higher scores mean more bot-like activity.

Contact: lynnetteng@cmu.edu

Description: BotBuster uses a mixture-of-experts approach to bot detection algorithm. This approach deals with incomplete data due to data collection limitations or account suspension. Each input (e.g. username, screen name, text) is trained individually with specific treatments to their quirks and separate predictions are performed corresponding to the available information. The predictions are then combined in a gating network to output a bot probability.

Contact: info@netanomics.com

Contact: info@netanomics.com

Description: BotHunter - A tiered Approach to Detection and Characterizing Automated Activity on Twitter.

Contact: huip@iu.edu

Description: BotSlayer is an application that helps track and detect potential manipulation of information spreading on Twitter. BotSlayer uses an anomaly detection algorithm to flag hashtags, links, accounts, and media that are trending and amplified in a coordinated fashion by likely bots. A Web dashboard lets users explore the tweets and accounts associated with suspicious campaigns via Twitter, visualize their spread via Hoaxy, and search related images and content on Google. 

Contact: psheth5@asu.edu

Description: CauseBox is a unified platform meant to serve as a benchmark for an ensemble of machine learning and deep learning based treatment effect estimation methods. It allows users to run and compare seven state of the art treatment effect estimation methods against benchmark datasets widely accepted in the causal inference literature. This tool is helpful for researchers who want to compare their own methods against benchmark methods. CauseBox supports GUI as well as command line interface.

Contact: mdeverna@iu.edu

Description: CoVaxxy is a web-based data visualization dashboard that allows users to concurrently explore the relationship between COVID-19 vaccine conversations, vaccine uptake, and epidemic trends in the United States. The dashboard tracks and quantifies credible information and misinformation narratives over time, as well as their sources and related popular keywords. Furthermore, vaccine uptake and conversation statistics are visualized geographically at the U.S. state-level. The dashboard is updated daily and the data that the dashboard utilizes is made publicly available for others to rehydrate via the Twitter API.

Contact: nlkolluri@utexas.edu

Description: Manual fact checking is unable to cope with the large volumes of COVID-19-related fake news that now exist. To help address the need to classify this fake news proliferation in the COVID-19 ‘infodemic’, we developed CoVerifi, an automated open-source tool to verify COVID-19 news and information. CoVerifi integrates crowdsourcing, newsfeeds, social media, and machine learning. Users of the web-based tool also have the ability to “vote” on news content, making the CoVerifi platform an effective method to collect labelled data. To develop our fake news detection tool, we built a crowdsourced dataset of ~7000 entries, which we tested and validated our CoVerifier model with. Ultimately, CoVerifi empowers users to make their unused consumption decisions by providing various points of data, rather than labeling news content as fake or fact. 

Contact: anubrata@utexas.edu

Description: Although we have seen a plethora of research in automated fake news detection, in practice, a large part of fake news detection efforts relies on human labor. Lack of adoption of fake news detection algorithms can be attributed to the complex nature of the problem and the high cost of error. We propose a tool that aims to close the gap by assisting human fact-checkers in their decision-making process. This tool adopts the methodology of evidence-based explainable fact-checking. Users can type a claim into a search box. With the help of search engines such as Bing, or Google, the tool retrieves a set of evidence relevant to the claim. Further, the tool calculates the stance of each piece of evidence and the reputation of each source. It aggregates the evidence to provide a veracity outcome. Users can also override model components (stance and reputation of the evidence) if the model makes a mistake. The model takes user input into account to update the claim veracity outcome.  ;  

Contact: bjiang14@asu.edu

Description: EXPO2O is a web-based dashboard that provides concise, intuitive, and interactive COVID-19 data visualization for users. Our dashboard allows users to visualize the potential relationship relations between various online-online, offline-offline, and online-offline data. EXPO2O also aims to improve interdisciplinary research on exploring relations between various types of data in a pandemic. In the demo, we will show preliminary findings and insights from the data we have collected so far: 

1. Online data: Google trends data, social media data, news media data;
2. Offline data: COVID-19 related statistics, US census data, local events/protests/policies.

Contact: ujeong1@asu.edu

Description: The purpose of this application is to provide an integrated data collection and analysis system for current research on fact-checking related to Facebook advertisements. Our system is capable of monitoring up-to-date Facebook ads and analyzing ads retrieved from Facebook Ads Library.

Contact: iem@andrew.cmu.edu

This tool constructs networks of Twitter users who are tweeting the same hashtags or same URLs within a short time window, in which tight clusters correspond to coordinated users.

Contact: juyheng@cs.cmu.edu

Description: The CASOS Hate Speech Detection model uses psycholinguistic features to predict the likelihood that a given tweet is hate speech. Due to multiple - and at times conflicting - definitions of hate speech, our model is trained on multiple datasets and produces likelihoods optimized for these different definitions. This affords users the ability to select the most appropriate predictions depending on their definition of choice, or run experiments with multiple definitions of hate speech for robustness. Across definitions, the model is trained using theoretically anchored features that allow for meaningful interpretations of results in relation to social identities and conflicts.

Hate Speech Detection:

Contact: torresch@indiana.edu

Description: Hoaxy provides an easy-to-use way to visualize the spread of information on Twitter. The user has the ability to visualize the spread of articles which they query from a list of URLs that are associated with fact-checking sources and low-credibility domains. Additionally, we leverage the Twitter API to allow users to visualize any search query that works on the Twitter search bar. Anatomy of an online misinformation network ()

Contact: iem@andrew.cmu.edu

Description: Image Emotion Classifier is makes use of machine learning to provide the probability of images identifying with each of Plutchik eight emotional categories: anger, fear, sadness, disgust, surprise, anticipation, trust and joy. It is trained using 8000 images tagged with the respective categories on Flickr. It has been applied in a case study to identify emotions in images in an emotional event - the Kashmir Black Day event.

Contact: team@lumendatabase.org

Description: Lumen is an independent research project studying cease and desist letters concerning online content. We collect and aggregate requests to remove material from the web. Initially focused on requests submitted under the United States’ Digital Millennium Copyright Act, Lumen's database, which offers API access for notice submitters and researchers, now includes complaints of all varieties, including those concerning trademark, defamation, and privacy, both domestic and international. Currently, the Lumen database contains approximately 17 million of removal requests, referencing 4.5 billion URLs, and grows by more than 20,000 notices per week, from companies such as Google, Twitter, YouTube, Wikipedia, Reddit, Medium, Github, Vimeo, and Wordpress.   

Complete API documentation is available at

Contact: kathleen.carley@cs.cmu.edu

Description: A network analysis toolkit for graphical, statistical and visual analytics on both social networks and high dimensional networks that can vary by time and/or space. ORA is a full function network analytics package that supports the user in creating, importing, exporting, manipulating, editing, analysing, comparing, contrasting, and forecasting changes in one or more networks. ORA pro can handle networks with millions of nodes and includes BEND analytics and a stance detector.  

ORA: A Toolkit for Dynamic Network Analysis and Visualization (pdf)  

Contact: kathleen.carley@cs.cmu.edu

Description: Computational linguistic tool for extracting semantic networks, meta-networks, sentiment and cues from texts and social media posts. Netmapper operated in over 40 languages. It also can extract phone numbers, emojis and emoticons. NetMapper User Guide v12 9/2021 (pdf).                                  

Contact: freelon@email.unc.edu

Description: PIEGraph is an interactive chart that displays web domains that have recently appeared in the user's Twitter feed in a scatterplot. The x-axis represents the domains' left-right ideological orientation, while the y-axis represents content credibility. The values for both axes were generated by the Media Bias Fact Check organization (https://mediabiasfactcheck.com/). The size of each bubble represents the relative prevalence of each domain--domains appearing more often have larger bubbles.

Contact: jessica.collier@austin.utexas.edu

Description: The Quiz Creator is a simple online tool using a step-by-step process to create a quiz in as little as 3 minutes. The interface is customizable to assist in seamless integration on any site. The quiz is embeddable to any page on a website and allows for tests of audience knowledge and response rate. Users can also create A/B tests to see which quizzes are most effective. We have done a series of projects to test the benefit of this tool: - ; ; and a working paper here:            

Contact: meganbrown@nyu.edu

Description: smaberta is a Python client for creating transformer-based classifiers in Python. Adapted from simple transformers, smaberta allows researchers to train, evaluate, and predict using minimal code.              

Contact: anwala@iu.edu

Description: StoryGraph quantifies the level of attention given to new stories by 17 US left, center, and right news media organizations. The service generates a news similarity graph every 10-minutes, where each news story is assigned an attention score indicating the magnitude of attention it receives collectively from the news media organizations.      

365 Dots in 2019: Quantifying Attention of News Sources ()

Contact: juyheng@cs.cmu.edu

Description: TrollHunter uses psycholinguistic features to predict the likelihood that a given account is a troll. Due to conflicting definitions of trolling, we opt for an empirically grounded operationalization that emphasizes the use of abusive, targeted language. TrollHunter relies on properties of not only the messages of the account of interest, but also any messages to which the account of interest may be interacting. This allows for context-aware predictions that align with our understanding of trolling as an interactive - and disruptive - phenomenon.       

Contact: kathleen.carley@cs.cmu.edu

Description: Construct, developed by CASOS, is a multi-agent model of network evolution. In Construct individuals and groups interact communicate, learn, and make decisions in a continuous cycle. The program takes into account how agents learn through interaction conducted over different media and change their information, beliefs, and activities based on what they learn. This can be used for forecasting how a network can evolve and seeing if two groups that appear identical on one dimension actually evolve in the same way. Training and Sample Data:

Contact: iem@andrew.cmu.edu

Description: Set of python scripts for collecting Twitter user timeline data, conversations, recent search tweets, full archive search tweets, sampled stream tweets, filtered stream tweets, and user profile information using the Twitter API v2. The scripts take care of the requests and writing out the collected data. Additionally, scripts are provided that convert the collected tweets from v2 format to v1 format so that they can be compatible for existing tools/software created for the v1 format of data. The v2 to v1 converter code can also be run in a standalone fashion for data collected through other methods.                                                 

Contact: meganbrown@nyu.edu

Description: This package makes working with link data from social media and webpages easier. It not only expands links, but catches errors, and makes parallel link expansion quick and efficient.

Contact: meganbrown@nyu.edu

Description: youtube-data-api is a Python client to download public YouTube data about channels, videos, and searches.