A new study by Rochester Institute of Technology and experts from Carnegie Mellon University challenges a largely owned belief about the media – that local media has become more critical of police in recent years.
Study, published in Proceedings of the National Academy of Sciences (PNA)used advanced artificial intelligence techniques (AI) to analyze the value of a decade of local reporting on the police across the United States. The team applied large language models to a set of data of more than 250,000 press articles of 10 politically diverse cities.
The researchers found that the criticism of the police in the local media had remained largely stable from 2013 to 2023. There was also little difference between the relationships in conservative and liberal cities – which means that local media did not adapt their report on the police to their audience’s policy.
Although the police ‘criticism peaks occurred after high -level incidents, such as the murder of George Floyd in 2020, the study revealed that these increases were temporary and did not lead to a long -term change towards more critical relationships. In fact, since 2020, local reports supporting the idea that the police are effective has increased slightly.
“The data told a different story about what we expected,” said ASIQUE KHUDABUKHSHDeputy Professor at Rit Golisano College of Computing and Information Sciences and co-author of the study. “Local journalism on the police has been stable, but the way people read the news depends on their political preference. Polarization is less news and more on readers.”
Perception vs reality in the consumption of news
This study is based on Previous research of the teamWho examined how the main national networks – MSNBC, Fox and CNN – covered police problems. This document revealed that cable news tends to direct the coverage to align with the expectations of the public, creating more partisan stories.
For this study, the researchers began by conducting an investigation. They found that a majority of people – a political affiliation, depending on whether the police reports have become more critical over time. However, the researchers have seen practically no empirical evidence on how police coverage has changed over time.
“Our goal was to look at a large temporal horizon – 10 years old – instead of focusing on a single moment in history,” said Khudabukhsh.
Provided
The plot shows how readers of different cities see news about the police. Democratic cities included Houston, Denver, Tampa, Nashville and Pittsburgh. Republican cities included Dallas-Fort Worth, San Diego, Jacksonville, Oklahoma City and Omaha.
To ensure balance, the study collected news from 209 points of sale in comparable cities – five in republican areas and five in democratic areas. Researchers applied natural language treatment methods to analyze 1.3 million extracts from local newspapers and television stations.
A key element was to ensure that various perspectives were incorporated into the AI system.
About 500 republicans, independent and real democrats read a sample of police extracts from the police to assess whether the coverage supported or contradicts two key declarations – “the police protect the citizens” and “the police is racist”. These human judgments have helped to train and refine the important language models that have measured the full police data.
“When building AI systems, it is essential to include a diversity of views,” said Khudabukhsh. “If we only authorize one political perspective to annotate data, we risk creating an unbalanced evaluation with biases. Each data point in our annotation study has been examined by a republican, a democrat and an independent. ”
This collaboration between Rit and Carnegie Mellon underlines how AI can be used to study large -scale media trends.
Provided
The researchers have formed an AI model to predict how the various readers would assess the sentences of the news on the police. The intrigue shows the percentage of sentences – each month – which were predicted as supporting and contradicting the hypotheses, “the police protected us” (on the left) and “the police is racist” (right). Some high -level reports are identified by points on the plot.
“It’s the best time to do this type of research because the AI landscape has changed and we have the tools and data to ask deeper questions,” said Sujan Dutta, a Informatics and information sciences Ph.D. Student at Rit and Co-author. “The biggest challenge was to present the notion of political perspectives in AI and make a very complete and balanced design which is representative of different points of view.”
The results of the study suggest that local journalists covering local problems are not prey to the type of partisanary observed at the national level. The researchers hope that their work offers a clearer and data on data from the way the police are covered in local news.
“The media informs a lot about our current state of the world,” said Khudabukhsh. “Better understand how sources cover news and people consume it, we can work to fill the polarizing gap that we have today.”