2014年3月24日月曜日

Lessons from Crisis Response in Japan: Finding Missing People

What we all know is that when natural disaster happens, people will be looking for their loved ones. At Google , we had been launching Google Person Finder for various crisis.

The big earthquake in Japan happened on 2011/3/11 at 14:46, the Googlers in Japan jumped on to translate the site, and 1 hour and 46 minutes later at 16:32 Person Finder was launched in Japanese local language.


However, translating the menu to local language alone will not solve many of the localization issues- so the team started to work on various enhancements.

1. Feature phones

Initial Person Finder was not usable from feature phones, and many people in Tohoku did not have smartphones, so this was crucial. Japan is a very mobile-savvy country, and hence there are many feature phones with various browser using various encoding, so coping with those differences needed local expertise.

2. Search

Although initial Person Finder was equipped with minimum internationalization, when people's names were more than 5 letters in Chinese character, it was not searchable. Also, different Chinese character have different pronunciation in Japanese, but pronunciation (yomigana) was not equipped as input nor in the search functions in the initial phase of Person Finder. In Japan, some people use different characters for same names- for example "渡辺" and "渡邉" are both used for the name "Watanabe" and search function needs to be able to cope with them. Making people searchable by phone numbers, making it searchable by romaji (alphabet), making it searchable by combination of name + address were all functions that were added in Japan based on people's needs.

3. Data conversion - analogue to digital

In the evacuation centers, information about people was shared on paper on walls. They were not digital, they were not online, nor searchable.


The team called out to take the pictures of those hand-written papers at evacuation centers with camera phones, and send those pictures by email which would be uploaded to photo-sharing site Picasa. Volunteers all over Japan (and sometimes outside Japan) helped to transcribe them.


Initially, transcribing by machine using OCR was tried, but since the handwritings were written in a hurry and not clean nor standardized, the quality was low. It needed to be done manually. So Google employees tried to transcribe them but there were overwhelmingly more photos than they could transcribe. Crowdsourcing was the best solution for this. External volunteers started to make rules, create lists of the photos that are not yet transcribed, and organically improved the manuals. This wiki was not created by Google- it was created collaboratively by the numerous volunteered that helped, instructing how to find photos that are not taken care of, how to announce they are working on specific photo (to avoid duplicate work) and how to announce it is finished, formatting of text, how to register to Person Finder database, how to double check the data, and how to triple check.


Over 5,000 volunteers helped to transcribe over 10,000 photos with over 140,000 records of people.

Lessons learned:
1) Adapt to reality - if people don't have Internet on the grounds, we have to be creative and adapt to the reality.
2) Build the platform- if you build the platform, people will be able to help voluntarily.
3) Trust the users and delegate
4) Unify communication methods - there were various mailing lists and communications in Japan's case- need to learn and unify next time.
5) Crowdsourced data input is a great way for people to be involved in helping -especially for people who cannot physically go and help and feeling guilty

4. Data migration

Person Finder data on Tohoku earthquake counted 670,000 records, which consists of 3 different data sources:
1) Data that users input to Person Finder manually
2) Data transcribed from the photos of evacuation centers, mentioned above
3) Data provided by mass media and other organizations
Mass media: NHK, which is the national broadcaster of Japan was calling out to provide whereabouts of people, and released those data on TV and radio but those data are not searchable. Mainichi News Paper provided the data from their reporters' memo from their visits to evacuation centers. Asahi News Paper had their data on the website, so Googlers wrote scripts to crawl those data.
Mobile carriers: Mobile phone companies operated their own database for people using their disaster BBS systems, which later became searchable via Person Finder, using PFIF - People Finder Interchange Format.
Police: Police was also gathering data on whereabouts of people, which later became searchable via Person Finder.
Local governments: Local governments had their own data of people, which later became searchable via Person Finder.

Lessons learned:
1) Standards, common format and interoperability is key
2) One "goto" site is important
3) Collaboration and preparation is needed before crisis happens ("Today" is the best time to start collaborating!)

One of my friends lives in Tokyo but her family is in Kamaishi, a town that was horribly struck by the tsunami. She once told me "Back then, mobile phone was dead and I couldn't reach them. We were not allowed to go to that area. I sent e-mail but they were not responding. There was no way for me to find out if my parents were dead or alive for several days. I was so worried that I was getting out of my mind. Then, someone put their names in Person Finder and I was able to find they were alive. I don't know how to thank Google Person Finder enough." I'm sure there are many more stories like this, but this is why we keep responding to disasters.

Disclaimer: The opinions expressed here are my own, and do not reflect those of my employer. -Fumi Yamazaki