The NewsReader team of the VU University Amsterdam organized a NewsReader hackathon at the Amsterdam public library on January 21st. Sound & Vision’s development team took part in the event. Its two main hosts Piek Vossen and Marieke van Erp introduced the NewsReader system and guided the attendants through the activities of the day.
INTRO: What is Newsreader?
Piek Vossen kicked off the day with an insighful presentation on the efforts behind processing millions of LexisNexis news articles to identify and store news events. The goal is to provide a specialized system for searching and visualizing coherent news story lines. Such a system could aid decision makers to gain quick insights into the reasons leading up to current events, such as a company takeover in the industry.
The dataset was optimized to find answers to the research question: How did the automotive industry change during the financial crisis?
Piek explained that using a document-centric search engine, such as Google, this question is hard to answer. The results from “Volkswagen + takeover”, for example, yield documents that contain these two words, but there is no way to quickly plot a chronological storyline about the when, where and who of Volkswagen-related business takeovers. The aim of the day was to give the participants the opportunity to find out whether NewsReaders’ event-centric data model indeed accommodates a better means to identify news stories.
Marieke van Erp then took the stage and continued with the practical details of the system. She explained how to use the API and, if you really felt like going all the way, its SPARQL end-point. Giving everybody a good basis to start experimenting, the VU team kindly provided a useful page with example queries and use cases.
After getting handed the tools, the Sound & Vision team (Jaap Blom and Themistoklis Karavellas) got together with Harold Kaperink and started hacking.
NewsReader video annotation
Our team worked on extending the editor tool (ET) developed within the LinkedTV project. This tool enables a user to search through different APIs and use retrieved links to annotate segments in a video. As Berlin news station RBB also participated as a use case partner within LinkedTV, we thought it was interesting to find a way to annotate RBB News materials with news events obtained from the NewsReader.
During the day, the team worked on how to use the NewsReader API to find interesting events related to news items available in the set of RBB news broadcasts.
Eventually we found a chapter on a corruption case involving the Berlin airport. After hooking up the NewsReader API to the editor tool we could use it to search the NewsReader API for related events, using, for instance, these politicians as input for our queries.
All that was hacked
No less than eight teams, including some solo teams, presented the fruits of their efforts. Sound & Vision’s team went on first. We outlined the context of the LinkedTV project and showed a live demo of the fully integrated NewsReader API within the editor tool.
Next up, several teams showed interesting statistics that came out of the dataset, such as: what universities are mentioned most in events concerning the car industry? Or by applying sentiment analysis: which car manufacturers are mentioned mostly in a positive way and which are prone to be the paraphrased in a more negative context?
The biggest team, an all-VU team, presented their attempt to enrich the NewsReader dataset with external sources such as Twitter and Google.
Award ceremony & wrap up
The winners were two one-man teams, who focused on either getting the most out of the dataset or on giving feedback on the performance of the API.
Our team really enjoyed the hackathon. It was well-organized and gave us the opportunity to learn more about event detection and event-based search; topics we at R&D are currently exploring in the DIVE project and are eager to experiment with in relation to our collection.
*Picture by NewsReader Team