The PI's team is working on a website for family reunification for the Haiti earthquake. In addition to crawling and scraping data from Web pages for the repository at http://haiticrisis.appspot.com/ coordinated by Google and many other volunteers, the team has built a powerful search interface on the data. During this effort, the team has identified several interesting research challenges to be studied in this project in order to make the system more scalable. The intellectual merits of this project include the support of powerful keyword search with efficient indexing structures and algorithms in a clouding-computing infrastructure that is increasingly popular for supporting large scale applications. The main challenge is how to use limited programming primitives in the cloud to implement index structures and search algorithms.
The techniques developed in this project will have a broad impact on many information systems that are moving to the cloud-computing paradigm. The adoption of the techniques in the family-reunification domain will have a significant impact in our society by helping people find their loved ones in a disaster. The PI's team will use the Google Person Finder project as a real application to test the techniques in the Haiti earthquake. The team plans to provide the techniques and source code in future releases so that the techniques can be used in family reunification during future disasters.
For further information see the project web site at the URL: http://fr.ics.uci.edu/haiti/
In this project we focused on how to support powerful people search for family reunification in disasters. We built a search prototype at http://fr.ics.uci.edu/ using our research techniques to support powerful instant fuzzy search for the Haiti earthquake (http://fr.ics.uci.edu/haiticrisis). We built a search prototype (http://haiti.rankfulltext.appspot.com/) using Google App Engine to support full-text search on people records. We studied how to improve GAE search capabilities using its limited interfaces. We studied various research challenges related to this new search paradigm in which the system can support instant search. We developed efficient indexing structures and algorithms to enable important features, such as error correction and location-based search. These techniques are very useful when users do not remember exact spellings of the answers, and/or need to find information related to a specific location. We studied how to use query log data of such systems to analyze user behaviors and quantify the benefits of instant-search systems compared to traditional systems. We compared the PSearch system (http://psearch.ics.uci.edu, which does instant fuzzy search on the UCI directory) with a traditional system (http://directory.uci.edu/). The results showed that queries to instant search systems have unique patterns, which can be used to better understand user search behaviors and improve system ranking. We also study how to improve ranking in this type of search by utilizing positional information of keywords.
