Safety Information Mining — What can NLP do in a disaster—

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 965–973, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Safety Information Mining — What can NLP do in a disaster — Graham Neubig Kyoto University Yoshida Honmachi, Sakyo-ku, Kyoto, Japan Yuichiroh Matsubayashi National Institute of Informatics Chiyoda-ku, Tokyo Masato Hagiwara, Koji Murakami Rakuten Institute of Technology New York Abstract This paper describes efforts of NLP re- searchers to create a system to aid the relief efforts during the 2011 East Japan Earthquake. Specifically, we created a system to mine information regarding the safety of people in the disaster-stricken area from Twitter, a massive yet highly unorganized information source. We de- scribe the large scale collaborative effort to rapidly create robust and effective sys- tems for word segmentation, named entity recognition, and tweet classification. As a result of our efforts, we were able to effec- tively deliver new information about the safety of over 100 people in the disaster- stricken area to a central repository for safety information. 1 Introduction On March 11, 2011 at 14:461, a massive earth- quake of Magnitude 9.0 struck off the coast of Japan. The earthquake and the ensuing Tsunami caused damage across the entire eastern coast of the country, with homes destroyed, roads impassi- ble, and large swaths of the disaster-stricken area losing electricity and the ability to communicate. Figure 1 shows the death toll and the number of missing people from tsunami and earthquake, presented by the Japanese National Police Agency. The number of missing people reached its peak on March 25th, indicating that the government took 2 weeks to fully grasp the total number of victims, although they began gathering information about the whereabouts of victims and survivors as soon as the earthquake hit. One of the main reasons for this delay was that prefectures could not effec- tively collect the information from municipal gov- ernments or police stations because many of them 1All the dates and times in this paper are in JST. 0 5,000 10,000 15,000 20,000 25,000 30,000 11-Mar 18-Mar 25-Mar 1-Apr 8-Apr 15-Apr 22-Apr 29-Apr 6-May 13-May 20-May 27-May The number of victims Date Dead Missing Total Figure 1: Change of overall death toll and missing people were extensively damaged or destroyed by the dis- aster (Anonymous, 2011)2. Furthermore, damage to communication equipment, power outages, and inundation of the air waves prevented the use of mobile phones, which are generally the most im- portant tool for communication during a disaster. As a result, Twitter or social network services (SNS) such as mixi3 played an important role for propagating safety information among peo- ple. However, with SNSs flooded with informa- tion about the disaster, it was difficult to ensure that people would be properly connected with the information they were seeking. Soon after the disaster struck, many NLP researchers, engineers, and students from all over Japan (including the authors), spontaneously created a working group called ”ANPI NLP”4 to try to organize this infor- mation into a …

Content truncated to 500 words