The NQS project I started at Stanford this year is part of a series of new initiatives aimed at tackling misinformation.
Last week, the Knight Prototype Fund awarded a total of $1 million to 20 projects aimed at improving the quality of information. The funders are the John S. and James L. Knight Foundation, the Democracy Fund, and the Rita Allen Foundation. The selection was made among 800 projects.
The winners include visual treatment of misinformation, media literacy programs, tools development, or fact checking powered by machine learning algorithms.
The full list of the projects supported by the Knight Prototype Fund is available here.
Among them is the project I worked on at Stanford’s JSK Fellowships called News Quality Scoring, NQS for short.
The following one minute video explains the project in broad strokes :
Here are the principles that support the NQS:
The production costs of good journalism and its economic value are currently uncorrelated because digital publishing tends to flatten everything.
I addressed these issues last October in a previous Monday Note.
By highlighting the most valuable part of the journalistic production, automatically and in real-time, we should be able to create a machine-readable signal of quality that advertisers and news distributors can take advantage of.
Just ten months ago, Facebook justified the elimination of its skeletal team of human editors saying editors didn’t scale anyway. As we all know, it proved to be a bad decision as Facebook was taken off-guard by the wave of fake news preceding the US presidential election. More than ever, scale matters. Facebook deals with 100 million links every day in 100+ languages, while many aggregators who operate at a much smaller scale still collect and distribute tens of thousands of stories on a daily basis.
The fates of publishers, distribution platforms, and advertisers are completely intertwined. However, they are in different phases of their business cycles. The business model for news publishing is in disarray. The digital advertising model is a failure for quality publishers with all its indicators flashing red. Large distribution platforms have captured most of the advertising without any windfall for publishers, luring them to one model, then switching to another as they see fit. Regulators (especially in Europe) try to do their job, but without any sense of anticipation nor technological acumen, they find themselves vulnerable to lobbies defending their own short term interests.
…need to highlight the most value-added part of their production. Even for noted news organizations, not all stories carry the same value. A large part of the news coverage is compulsory as it is part of the “basic” feed. A smaller chunk is the result of a genuine journalistic effort, whether it entails assigning several reporters to a story, sending someone in the field, or maintaining a network of correspondents and specialists. As of today, there is no automated system able to skim these types of content, whether it is text, video, visual stories, or data-driven treatments. These elements should be able to carry a higher advertising value and/or be part of premium subscription packages. Selecting quality stories could also vastly improve the relevance of recommendation engines and user engagement.
…will also be able to enhance the value of their inventory and reinforce audience loyalty. They, too, have an economic interest in separating the chaff from the wheat: Currently, their income relies on super-low CPMs based on the lowest common denominator of their stream of news.
…could also leverage NQS to increase their revenue by raising the price of ads displayed next to a content deemed qualitative. When presenting the NQS project to multiple interlocutors over the last months, I’m always asked the same question: Why should advertisers and media buyers be interested in any notion of quality? Here are my answers:
- High-end advertisers prefer quality context. Brands (especially luxury ones) are increasingly concerned by what is published next to their ads.
- Quality content calls for higher demographics with more educated, more affluent readers — both groups being more willing to pay for news content.
- Quality leads to higher viewability of ads: Visitors who read for more than 75 seconds see more than 60% of advertisements, according to Chartbeat.
- Quality editorial is a great fit for bespoke, high-value content advertising. The New York Times, The Washington Post, and Quartz have created their most profitable ad products thanks to the proximity of great journalistic content.
- Lifting quality from the noise could lead to the creation of premium ad networks for distribution platforms.
Finally, the public will also benefit from a cleaner editorial environment as NQS should contribute to making news quality more visible and accessible while reducing the amount of fake news. Also, by assigning a quality score to the entire editorial chain, NQS will be able to triangulate on suspicious news providers.
- The News Quality Scoring Project has been created during my time as a JSK Fellow. I expect my advisors at the JSK as well as the Department of Communication at Stanford University to remain closely involved.
- The French startup Kynapse.fr will be the lead technical partner for the NQS Project. It will provide most of the heavy lifting in machine learning and natural language processing. This company, created less than eighteen months ago, is specialized in large AI projects, quantitative marketing, fraud detection, and “smartbots” implementation. Its clients includes energy, health, and transportation conglomerates.
- NewsRepublic is one of the most successful Android news aggregators. Every hour, it processes 3,000 pieces of news, in six languages from 2,700 sources. Created by the French entrepreneur Gilles Raymond, it was acquired last year by the Chinese giant Cheetah Mobile which has 600 million active users. NewsRepublic will be involved in the creation of the initial dataset that will be used as the primary corpus of news content to be filtered and analyzed.
- The Donald Reynolds Journalism Institute at the University of Missouri is helping the project in building some key database of references.
- Storyzy is a French company specialized in natural language processing. It recently rolled out a quote verifier to tackle fake news.
In the fall, I expect to announce other key supports, including an agreement with a major US-based ad mobile platform.