Project: Active Collection of Structured Situational Knowledge via Symbiotic Human-Machine Computing
Summary: In order to incorporate the vast amounts of knowledge available from social media into Common Operating Pictures for disaster management, data streams need to be sorted, clustered and reasoned at close to real-time. To perform this extremely large and complex task we are investigating a new paradigm for collaborative intelligence call Symbiotic Human-Machine Computing which combines active machine learning with human-based computation to perform large-scale knowledge acquisition, information processing and actuation of events in the real world.
During life threatening situations, such as natural and man-made disasters, the initial goal of an emergency response team is to reduce the number of causalities by performing rapid situation assessment and directing the right resources to the right places to stem further damage and loss of life. This obviously requires the rapid collection and analysis of large amounts of disparate information about post-disaster conditions of infrastructures and components. Social media enables citizens to almost continuously report about disasters providing unparalleled speed for gathering information relevant for situational awareness. For example, within an hour after the 8.9 magnitude earthquake in Japan, 1200 tweets per minute were generated from Tokyo via Twitter and around 2000 users provided information for Haiti earthquake in a two weeks time frame. The information provided through social media is unstructured in nature due to its spontaneity and the fact it arrives at varying frequencies and granularities. In order to incorporate this vast amount of knowledge into Common Operating Pictures for disaster management, these data streams need to be sorted, clustered and reasoned within spatial and temporal dimensions and information must be extracted for use by emergency response teams. The usefulness of these reporting sources however is limited as there is a lack of tools to efficiently filter, classify and parse critical information in an accurate and timely fashion.
To solve these problems we propose a new computational architecture we call Symbiotic Human-Machine Computing. This methodology combines machine intelligence with human-based computation to capitalize on distributed human knowledge and observations. Human intelligence cannot be matched, in terms of its ability to rapidly learn and adapt to new situations and itâ€™s ability to robustly understand and infer knowledge. These are all inherent weaknesses in current machine learning approaches. We propose to extend on earlier works in human-based computing by developing symbiotic systems that combine, human observation, active machine learning and human-based computing to realize intelligent symbiotic systems that are capable of large-scale knowledge acquisition, information processing and actuation of events in the real world.
The problem of collating structured situational knowledge from diverse, dynamic and unstructured data sources require such a computational framework. Due to the complexity, diversity and the socially dynamic nature of natural language, purely computational approaches will be inadequate. Human-based computing is one possible solution however it alone is inadequate as the delay incurred to process large amounts of data are typically very large. We believe Symbiotic Human-Machine Computing is the appropriate solution for this problem. By leveraging human-based computing to bootstrap and improve computational systems for knowledge extraction in real-time, highly accurate and low-latency processing of incoming data can be realized. In addition to pervasive knowledge acquisition, the proposed framework will also be extended to enable computational systems to actively interact with human agents by way of requests for clarification, requests to validate information and requests for reporting of missing or unreported knowledge.
For more information, contact firstname.lastname@example.org.