The LMI data science team of Hasan Khan, Nathan Shepherd, Dominic Thomas, and Justin Ward wanted to tackle a real-world intelligence problem in its submission to the EPIC App Challenge. The second annual competition, hosted by AFCEA International’s Emerging Professionals in Intelligence Committee (EPIC), tasked competitors to develop a tool that extracts, enriches, and curates intelligence from publicly available data sources.
 

“Instead of attempting to find intelligence content in the news, we built a system to monitor newspaper sentiments for specific entities and how those sentiments change over time,” Hasan said. Knowing when opinion shifts occur is highly sought by open-source intelligence analysts, who track everything from large geopolitical developments to the domestic popularity of foreign leaders. “This could give more context to information analysts already have.”
 

Hasan and his colleagues, assisted by Lucas McCabe, developed a sentiment-monitoring system that depicts the number of positive, negative, and neutral articles published about a specific entity over time. The system resonated with the judges, all national security professionals in the private sector. LMI placed third, receiving a $1,000 prize, which was announced at the Intelligence & National Security Summit on September 4.

The competition furnished a meaningful opportunity for team members to strengthen their skills, particularly working with natural language processing (NLP), and begin exploring national security applications for their work. “Before this, I had zero background in NLP," said Nathan, who, like his colleagues, has primarily supported clients in LMI’s defense and health/civilian markets. “It’s been very helpful diving into [NLP tools]. I learned a lot about how these tools can be applied in the national security space and beyond.”
 

The month-long development process started with All the News, a publicly available dataset containing articles from 15 popular English newspapers from 2016 to 2017. Article texts and metadata were extracted, organized, and cleaned in the Python programming language. Sentiments were then calculated for each article using sentiment analysis libraries, including VADER (Valence Aware Dictionary and sEntiment Reasoner), StanfordNLP, and TextBlob. The negative, positive and neutral article counts for each publication source were aggregated by month and visualized as a line graph.
 

“We are incredibly proud of this team, one of the few in the competition not to borrow from an existing proprietary technology. The platform was designed and constructed from the ground up, burnishing LMI’s reputation as a leader in innovation,” said Brant Horio, director of data science. “The team’s submission deserved this recognition, which helps strategically position LMI for future national security work.”