Identifying Misinformation on Twitter: Analysis of Boston Marathon Bombings

• The text discusses the importance of trust and credibility in online social media, particularly focusing on analyzing misinformation and rumors.
• The analysis includes examining data from events like the Boston Marathon bombings to study how information spreads on platforms like Twitter.
• A graph illustrates the volume of tweets over time after the Boston blast, showing a distinction between legitimate information (in red) and rumors (in blue and green).
• Examples of fake information spread during the Boston blast event are provided, highlighting the impact of false tweets being retweeted thousands of times.
• Strategies are discussed to address the spread of false information on social media, including identifying rumors early and reducing their propagation.
• Techniques for identifying false information involve analyzing user features (e.g., number of followers, verification status) and tweet features (e.g., length, emoticons).
• Classification techniques like Naive Bayes and decision trees are applied to classify posts as legitimate or fake, with decision trees showing higher efficiency.
• The study involves analyzing a large dataset of 7.8 million tweets related to the Boston Marathon event, with millions of users posting, geo-tagging, and engaging with the content.
• The methodology used to collect data from events like the Boston Marathon involves identifying relevant keywords and hashtags to track and analyze social media posts.
• Geotagging data reveals the distribution of tweets from different locations, with a concentration in the US during the Boston Marathon event.
• The analysis includes studying spikes in tweet volume post-event and examining how social media content correlates with real-world incidents.
• The goal is to differentiate between true and fake posts by analyzing features like tweet content, user profiles, and engagement metrics.Here is the summary of the text:

• The study analyzed tweets related to the Boston Marathon event to identify rumors and fake news. • 29% of tweets were found to be rumors, and 20% of tweets were generated by fake accounts. • The study collected 32,000 new Twitter accounts created during the event, but 90% of them were deleted or suspended. • The analysis showed that fake accounts were highly connected and operated in a closed community. • The study used machine learning approaches to identify legitimate and fake posts, including feature extraction, human annotation, and model creation. • The annotation process involved classifying tweets into four categories: contains information, related to the event, not related to the event, and skip. • The study used a metric called NDCG (Normalized Discounted Cumulative Gain) to evaluate the efficiency of the classification model. • The analysis found that features such as user information, tweet content, and recency of tweets were important in determining the credibility of a post. • A Chrome browser extension called Tweet Cred was built to identify credible and non-credible tweets in real-time using the developed model. • The extension uses 45 features, including presence of swear words, negative emotion words, and web of trust scores, to make judgments on tweet credibility. • The top 10 features that influence tweet credibility include number of characters, unique characters, and words in the tweet, as well as user information and tweet engagement metrics. • The study highlights the importance of analyzing social media content to identify credible and non-credible information, particularly in emergency response situations. • The analysis techniques used in this study can be applied to other social media platforms and domains beyond Twitter.