IBM Watson Natural Language Processing technology powers Weather Channel COVID-19 Incident Map

By and Laura Chiticariu | 3 minute read | June 16, 2020

Misinformation, disinformation and a lack of fact-checked information have interfered with the global fight against COVID-19. In March 2020, when COVID-19 began to rapidly spread globally, it became clear that the world population could benefit from the data analysis and processing performed by Data and Artificial Intelligence technologies. Seth Dobrin, IBM Chief Data Officer, specifically noted that global citizens lacked a local source of current and accurate COVID-19 information, including trend rates, confirmed cases, and deaths.

This week’s Natural Language Processing (NLP) blog will center on the IBM Data and AI team resolution to consolidate state-level and county-level information across 50 state websites, and local government websites around the United States. Seth assembled a team with members from IBM Research (like myself, Laura Chiticariu), IBM Watson, and IBM Data & Governance to create a clean, easily digestible information source that would enable people to understand COVID-19 activity in their communities. The IBM cloud infrastructure would be essential to scale this service to millions of interactive users.

Using the Weather Channel website and app as a platform, the COVID-19 incident map is a data dashboard available to current app users and the greater public. The Weather Channel has compiled data provided by third party sources including the World Health Organization, state and local agencies and others. Access to this relevant and accurate information, is intended to help individuals make the best decisions, understand the risks in their own community, and ultimately stop the spread of COVID-19.

Powered by Watson NLP

To populate the Interactive Incidents Map, the latest and most up-to-date information must be presented in a standardized format in near real time. Accomplishing this task was not simple, and it was not possible without IBM Natural Language Processing.

The heart of the technology challenge was that county and state-level data sources do not use a standard API with structured data. There is no enforced technology format standard. County and state-level COVID-19 data is rich and stored in heterogeneous forms including unstructured text, semi-structured web pages (HTML), PDF documents and images.

This was a job for NLP. Watson NLP could parse and extract information from the complex data sources, and understand the information like a human.

To understand and extract the data necessary to feed the pipeline and ultimately the maps, the following Watson Natural Language Processing technologies were used:

  • Watson Natural Language Understanding, to extract insights from natural language text
  • Watson Discovery, specifically the Smart Document Understanding table understanding and PDF conversion features, to extract insights and answer questions from a variety of business content, such as PDFs, HTML, tables and images.

Today, with IBM Watson Natural Language Processing, the Weather Channel COVID-19 Incident Map provides localized reports in the United States down to the zip code. There are 41,701 zip codes in the United States.

Enacting Action

The solution went live on the Weather Channel website and app on the evening of Monday, March 30. Within 3 hours on Monday night, a million unique users had visited the platform. The platform hosted another 3.5 million users on Tuesday, and another 6 million on Wednesday. Now over 300 million monthly visitors are using this solution to access county-specific COVID-19 information.

The site automatically detects the user’s geographic coordinates to deliver trusted data from governments, the World Health Organization (WHO), and the Center for Disease Control (CDC).

Since deploying this solution, the team has seen interest from government officials, hospitals, and supply chain logistics organizations. These leaders are interested in using this tool to analyze trend curves in local communities and redeploy resources to communities in need.

Created by a Unified IBM team

During COVID-19, IBM has adopted social distancing and work from home tactics to protect the public and its employees. Sixty decentralized workers across different IBM teams created the COVID-19 Incident Map. This project was completed from initial ideation through to testing and deployment in 7 days.