Table of Contents
AI-Powered Recommendation Engine Redefines News Sourcing
AI-Powered Recommendation Engine Redefines News Sourcing – In the ever-evolving landscape of journalism, the quest for accurate and compelling news hinges on credible sources. Imagine a news story devoid of precise origins—unthinkable, right? Journalists grapple with the challenge of sifting through an abundance of potential interview candidates, each with their own biases and expertise.
Enter artificial intelligence, the guiding force behind a groundbreaking tool developed by researchers at the USC Information Sciences Institute. This innovative machine, a source recommendation engine, aims to assist journalists in finding relevant references by cross-referencing a vast database of potential sources, experts, and information.
Emilio Ferrara, a computer science and communication professor at USC Viterbi School of Engineering, explains, “In practice, the software application analyzes specific texts or topics and suggests relevant sources by cross-referencing against a database of potential interviewees, experts, or information sources.” The tool provides contact details, areas of expertise, and the previous work of potential sources.
The brain behind this tool is Alexander Spangher, a Ph.D. student at USC Viterbi and former data scientist at The New York Times. Spangher, having witnessed the pressures of traditional newsrooms, emphasizes the need to aid overwhelmed journalists in the face of an overwhelming volume of news.
To create the AI model capable of source recommendations, researchers first laid the foundation by studying how human journalists currently utilize sources in news writing. They analyzed over a thousand news articles, annotating information sources and their categories (e.g., “direct quotes,” “indirect quotes,” “published works,” and “court processes”).
Although annotating a thousand news articles wasn’t sufficient, the researchers used a language model (LM) to continue the annotation process. Ferrara clarifies, “Language models are AI frameworks that process and understand human language by analyzing large amounts of text to discern patterns and context.”
The trained LM achieved an 83% accuracy in detecting source attributions, allowing researchers to annotate around 10,000 news articles. The AI model unveiled insights into the composition of news writing, revealing that about half of the information in articles originates from sources. Furthermore, it found that articles typically feature one or two main sources contributing the majority of information, with two to eight smaller sources providing additional details.
The AI model also discovered that the first and last sentences of an article are most likely to be sourced. Spangher notes that reporters often begin with quoted information and end with a quote to captivate readers.
In a test challenging their algorithm, researchers investigated whether the AI could detect missing sources. Analyzing 40,000 articles with randomly removed sources, the AI effortlessly identified when a primary source was absent but struggled with smaller sources. While these sources may seem less critical, they could offer valuable recommendations in the future.
Spangher emphasizes, “You’ll get a lot of information from the main players, but additional voices can add color and additional details to the story. It will be a challenge to make the machine recognize and recommend smaller sources, but they may be the most helpful.”
Researchers envision the tool’s significance in recommending diverse sources, introducing journalists to new and varied perspectives beyond their usual networks. However, they acknowledge the inherent vulnerability of any AI system to bias if not meticulously designed.
Jonathan May, a computer science research professor at USC Viterbi and a lead researcher at ISI, envisions a future where source machines accelerate the reporting process, making journalists more efficient. May says, “Using technology to support our artistic endeavors is also a creative endeavor in and of itself. That’s the reason I have hope for it.”
The group intends to work with reporters to obtain input for future enhancements. Spangher expresses enthusiasm for engaging with journalists, understanding their needs, perspectives, and what they believe will succeed or fail. He emphasizes that any solution for local journalism requires collaboration from diverse individuals with different backgrounds.
The AI-powered source recommendation engine developed by the USC Information Sciences Institute marks a transformative leap in the realm of journalism. By providing journalists with a tool that can efficiently sift through vast amounts of information, identify missing sources, and recommend diverse perspectives, the potential for enriched and well-rounded news reporting becomes evident. As the research team plans to collaborate with journalists for further refinement, the future promises an integration of artificial intelligence that not only streamlines the creative process for journalists but also enhances the depth and breadth of news narratives.
This innovative endeavor stands as a testament to the power of technology when harnessed for creative solutions. As the AI source recommendation tool evolves, it has the potential to redefine journalistic practices, introducing efficiency without compromising the integrity of reporting. With a commitment to addressing biases and a collaborative approach to improvement, this development opens up new horizons for journalism, fostering a future where man and machine work in tandem to deliver insightful, diverse, and impactful news stories.