ABOUT SWIFT RIVER

Introduction

User-generated content is becoming an increasingly important source of information during emergency events while traditional media continues to play a pivotal role in documenting events as they unfold. These trends are expected to continue well into the future. The challenge, then, becomes filtering this growing torrent of information. There is an apparent tradeoff between crowdsourcing (opening the floodgates) and validation (the filter). One of the strengths of crowdsourcing is the ability to collect a high volume of information from highly diverse channels like Twitter, email, news sites, blogs, and SMS.

Swift acts as the verifying filter for these different channels and is possible precisely because of the volume of information available from these sources. The more information generated, the more the community interacts with it, and the easier it becomes to identify mutually trusted sources.

How Does it Work?

Although the general concept is simple, Swift relies on three incredibly complex technologies: Natural Language Computation, Machine Learning and Veracity Algorithms. The sum of these parts allows an emergency response organization to track and verify the accuracy of reports during a crisis, or a team of journalists might use Swift to track specific topics they happen to be researching. Swift helps surface authoritative sources, while suppressing noise (like duplicate content, irrelevant cross-chatter and inaccuracies.)

But how? When users begin monitoring a topic, they enter several related feeds. Swift begins aggregating these feeds, taking multiple channels, mashing them together and outputting one unified feed. To clarify, the user may be tracking Twitter, various Blogs, as well as a dedicated phone number (SMS), email, and news media. Swift then mashes those differing channels together into one feed, keeps track of where each item originated (it's source) and assigns a score to each source. This score is determined partly through user behavior and partly by our algorithms.

Meanwhile, our natural language computation service SiLCC, is used for what's called 'predictive tagging'. This is important because the act of tagging content is tedious and humans will often fail to do it. By using this natural language service, Swift can examine content and extract the keywords that are most relevant. For the headline "Major Earthquake in Chile", the important keywords are going to be 'earthquake' and 'chile'. These keywords can be used to find other content referencing the terms 'earthquake' or 'chile'. In essence that is 'predictive tagging', where algorithms try to extract meaning to help improve sorting. Users can vote on these tags to help our algorithms improve. Using these features, Swift users can determine the relevance of and relationship between content, regardless of the source.