Wellphoto / Istock
What would be the news without appropriate sources? To tell a convincing story, journalists must find accounts worthy of information and trustworthy information. This information generally comes from a large pool of publications, official documents and experts, all with their own prejudices, expertise, opinions and horizons. The interview candidates are abundant but overwhelming to sail.
Artificial intelligence, however, can serve as a guide.
Researchers from USC Information Sciences Institute Create a source recommendation engine designed to suggest references for journalists. “In practice, the software application would analyze a given text or subject would suggest relevant sources in cross references against a database of potential interviews, experts or information resources,” said Emilio FerraraIT and communication teacher at USC Viterbi engineering school. “The tool could provide the contact details, areas of expertise and the previous work of sources,” he added.
The development of the tool is led by Alexander StereA doctorate in computer science. USC Viterbi student who previously worked as a scientist of data to New York Times. Although plunged into the journalism industry, a steher has witnessed the pressure of traditional editorial rooms. “I did not speak to a single local journalist who was not completely exaggerated,” he said. “There were deserts of news and papers that closed. These are areas like this that we really want to help and build tools. “”
Motivated to provide useful resources to journalists, a member created various AI gadgets, including a prefaced source recommendation system in its article, “Identify sources of information in press articlesWhich was accepted for the 2023 Conference on empirical methods in natural language treatment.
To create an AI model that can suggest sources, researchers first laid the foundations: how do human journalists currently use sources in news writing? To study this, they gathered a set of sentences of more than a thousand press articles and annotated the source of information, as well as the supply category (for example, the “direct quotes”, the “Indirect quotes”, “published works” and “legal proceedings”).
However, a thousand annotated press articles were not enough for researchers to draw firm conclusions on all the countless ways in which journalists use sources in the kinds of reporting. However, it was enough to form a language model (LM) to continue the annotation process. “Linguistic models are AI frameworks who deal with and understand human language by analyzing large volumes of text for models and context,” said Ferrara, the main author of the article.
The LMS trained by the researchers could detect source attributions with an 83%precision, the authors revealed. Now equipped with these LMS, they have annotated around 10,000 press articles and explained more understanding of the compositionality of news editorial staff: when and how do journalists currently use sources?
AI models revealed that in average, about half of the information in press articles came from sources and, in each article, there is generally one to two major sources (that is to say these CI contribute to 20% or more information from the article) and two to eightsminers (those which contribute less). “The AI also discovered that the first and last sentences were the most likely to be obtained,” said stere, adding that journalists often lead with information cited and end with a quote to send the reader.
Could the researchers challenged their new algorithm with another test: could they detect if a source lacked? If AI can recognize when the information is lacking, it can be configured to find out when recommending a particular expert to complete the full image.
By analyzing 40,000 articles with certain sources removed at random, AI models easily noticed when a major source was absent but had difficulties with minor difficulties. Although they can be the least crucial for a story, less obvious sources can also be the most precious recommendations that AI could one day make, said stere.
“You will draw a lot of information from the main participants, but additional voices will provide additional colors and details to the article,” he noted. “It will be a challenge for the engine to recognize and recommend minor sources, but they can be the most useful.”
Researchers also think that the tool will be important if it can recommend various sources. “He can present to journalists new varying voices beyond their usual network, thus reducing dependence on familiar sources and potentially bring new perspectives,” said Ferrara.
However, each AI system is subject to biases if they are not appropriately designed, he added. “To ensure the diversity of source databases, standards should include the representation of a wide range of demographic data, disciplines and perspectives,” he noted.
Jonathan MayAn Associate Professor of Computer Science Research at USC Viterbi and Senior ISI Researcher ISI, imagines a future where the supply engine relaunches the report process, allowing journalists to be more effective.
“The technology that can help us do creative work and be our best creative is a good thing,” said May, co-author of the newspaper. “That’s why I hoped.”
The team plans to collaborate with journalists to collect comments for new improvements.
“With projects like this, I really flourish to speak to journalists and understand their needs, their views and what they think or will not work,” said Spangher. “Any solution to local journalism will require a lot of different people with a lot of different backgrounds.”
Posted December 6, 2023
Last update on May 16, 2024