Paper Reading #25 :TwitInfo: Aggregating and Visualizing Microblogs for Event Exploration

Reference

Authors: Adam Marcus, Michael S. Bernstein, Osama Badar, David R. Karger, Samuel Madden, Robert C. Miller

Affiliation: MIT CSAIL,32 Vassar St., Cambridge MA

Presentation: CHI 2011, May 7–12, 2011, Vancouver, BC, Canada.

Summary

Hypothesis

For people trying to understand events by querying services like Twitter, a chronological log of posts makes it very difficult to get a detailed understanding of an event.This paper presents TwitInfo, a system for visualizing and summarizing events on Twitter.

Contents

TwitInfo allows users to browse a large collection of tweets using a timeline-based display that highlights peaks of high tweet activity. A novel streaming algorithm automatically discovers these peaks and labels them meaningfully using text from the tweets. Users can drill down to subevents, and explore further via geolocation, sentiment, and popular URLs. They present an algorithm and an user interface as the TwitInfo system.

An evaluation of the TwitInfo system revealed that users were able to reconstruct meaningful summaries of events in a small amount of time.

Methods

To evaluate the user interface, 12 participants , 6 with twitter account were asked to do searches that would use the most of the features on interface.

To evaluate the algorithm, the researchers gathered tweets from three soccer games and one month of earthquakes. To create ground truth for the soccer data, one researcher annotated major events in the soccer game using the game video and web-based game summaries, without looking at tweets. For ground truth on the earthquake data, they gathered data from the US Geological Survey on major earthquakes during the time period.

Then participants were given five minutes to understand an event using TwitInfo and five minutes to dictate a news report on the event to the experimenter. Later an interview was conducted to get qualitative data.

Results

When all earthquake related discussion were included, precision rose to 89% and similarly it leads 95% precision in the soccer dataset while all related discussions are inlcuded.

The algorithm has high recall, finding all major earthquakes and most soccer events.

The algorithm can be biased by Twitter’s interests.Most users relied on tweets to confirm event details, though a few skimmed the automatically generated labels. Under time pressure, people skimmed all peak labels to get a broad sense. Timeline and event labels were noted as the most memorable and helpful elements.Participants were successfully able to construct the details about an event based on the TwitInfo timeline. People wanted more information and aggregation on maps. People were not so much impressed with the performance of sentiment analysis model and were concerned about which results they need to pay attention to.

Discussion

The application that the researchers developed was pretty fascinating to me. I tried to find if there was an application for general use but was not able to do so. I would have loved to have a desktop or a mobile application to play with and see how it performs. If two important events are happening at the same place and people are tweeting about both of them, and if the application could let us evaluate both of them at the same interface then the application would be even better. If the application can as well provide functions to filter tweets from individual places or let people define constraints on locations of tweets, that information can help users make better assumptions and conclusions. For example, for the game between liverpool and manchester city, the emotions might have been quite opposite at two different places.