News

Best practices for verifying UGC photos

The process of verifying user-generated content (UGC), particularly photos, has long been “more art than science,” as NPR’s Andy Carvin, who curates UGC from the Middle East on his 96,000-follower Twitter feed, said in 2011. But a recent study was able to identify fake photos with 97 percent accuracy, suggesting that technology will soon allow journalists to streamline the UGC verification process to the click of a button.

by WAN-IFRA Staff executivenews@wan-ifra.org | July 8, 2013

Research by the Indraprastha Institute of Information Technology in Delhi, India and IBM Research Labs analyzed 10,350 fake image Tweets from Hurricane Sandy. Such photos, as the Atlantic’s Alex Madrigal pointed out, generally come in three varieties: photoshopped pictures, falsely-marked old photos resurfaced from previous events and anachronistic, photoshopped images. As photos of Hurricane Sandy flooded the web at a rate of 10 Instagram photos per second, journalists such as Madrigal took on the roles of a “pictoral investigation bureau” to separate legitimate UGC from the deluge of fake imagesBuzzFeed, the Guardian and Tom Phillips’ “Is Twitter Wrong?” Tumblr, among others, also monitored fake Sandy images.

Such UGC verification is particularly important during disasters, as the climate of a crisis situation minimizes the importance of a source’s Twitter reputation, so fake accounts’ content can go viral. During an event like Hurricane Sandy, people increasingly rely on Twitter’s search function rather than just their timelines, according to Poynter. This allows fake or doctored photos to echo far beyond the source’s follower base. And just one retweet from a credible source can allow a fake photo to spread rapidly. The study found that 86 percent of all the tweets with fake photos were retweets.

A decision tree classifier algorithm, which examined user features, such as the source’s number of followers and biographical information, and tweet content, such as length, type of punctuation and presence and type of emoticons, allowed the researchers to identify fake photos. The hope is that this science will eventually be rolled out for everyday use. The researchers mention developing a browser add-on that could automatically detect fake Twitter images, Poynter reported.

The AP uses similar software by Hany Farid, a researcher at Dartmouth College, said Santiago Lyon, vice president and director of photographer at AP. Lyon acknowledged, however, that such software is imperfect and “cannot yet detect every skillful manipulation.”

As this technology is still being tested and perfected, journalists must rely on other means to confirm UGC. But “journalistic hunches” can be fallible, as BuzzFeed’s quiz of real or fake Sandy photos shows. Despite the sense of 24-hour news cycle’s urgency, it’s essential to make sure all user-generated content is authenticated before publishing it.

“Even if something is incredibly compelling and it doesn’t pass one of our steps, then it doesn’t go out,” Fergus BellAP’s Social Media & UGC Editor, International, told Poynter. “That’s how we stop from being wrong, which is tough sometimes, especially when it’s something that’s really great. But we just don’t put it out, because the [verification] system has grown organically and it hasn’t failed us yet, and so we trust it.”

It’s important to establish a defined step-by-step process for validating UGC, as doing so will ensure both speed and accuracy, Poynter reported. What follows is a list of best practices and tips for verifying and using UGC, based on policies of media organisations including the BBCAP and Storyful.

  • Assume all UGC is fake until presented with enough counter-evidence, as is the Storyful mantra. Merely quoting social media content and attributing it to an online username is not acceptable, according to AP guidelines.

  • Seek the original source of the information. TinEye Reverse Image Search and Advanced Google images search allows you to pull up a photo’s other appearances online, Storyful’s Fiona McCann wrote. If the source wasn’t the initial provider of the content, it’s important to identify him or her and ask for permission to use the content, according to UGC Alliance.

  • Piece together a background of the source using all information available. “Almost all your contributors will have some kind of online footprint,” said Mark Jones, the global community editor of Reuters. The source’s Twitter bio and other associated accounts can be used not only to validate their claims but also to determine whether they have a compromising stake in the issue. Red flags for fake UGC include a low number of tweets and newly-created Twitter accounts, according to iRevolution. Another sign to look for is tweet spam, when a source repeatedly posts the same image or link, the Guardian reported.

  • Establish direct contact with the source. The “golden rule” for UGC verification: speak on the phone to the source, staffers of BBC’s UGC Hub told Nieman Reports. Not only will this process give the reporter more context on the photo, the response can provide valuable information on the source. “Even the process of setting up the conversation can speak volumes about the source’s credibility: unless sources are activists living in a dictatorship who must remain anonymous to protect their lives, people who are genuine witnesses to events are usually eager to talk,” David Turner wrote.

  • Look past the foreground of photographs. Some images include metadata that provides information about when and where they were taken, said Lila King, participation director for CNN digital. However, only the original, high-resolution file will provide this level of detail, according to Nieman Reports. But even without this data, details such as shadows, foliage, license plate registrations, signs and weather can help confirm the reported time and location of the photo, according to Storyful. In the case of video, the BBC also works with language specialists to ensure the dialect and accent match the location of filming, BBC staffer Alex Murray said.

  • Cross check content. It’s possible to use webcams to survey the area of the photo, the Guardian suggested. The AP contacts local bureaus to confirm weather or specific geographic information, according to Poynter.

  • Properly disclose UGC. The AP adds disclaimers that acknowledge no reporters were able to “independently verify the authenticity, content, location or date of this handout photo/video” when direct communication with the source is not possible, according to Nieman Reports.

  • Credit your sources. “On-screen credits or caption credits are becoming much more common across the industry, and I think that’s a real sign that people are recognising that these people deserve that credit and they’re not just someone randomly uploading a video or uploaded a photo necessarily,” Bell told Journalism.co.uk.

Share via
Copy link