An inside look at Google's news-ranking algorithm
- 21 February, 2013 19:42
A patent application filed by Google last year provides a detailed look at some of the metrics the company considers when ranking news stories and deciding how prominently to display them on its Google News page.
The application, filed in February 2012 and published last July, seeks to build on a patent Google was issued in 2009 titled "Systems and Methods for Improving the Ranking of News Articles." Computerworld found the document while conducting an unrelated patent search on the United States Patent Office's website.
A Google spokesman had no comment on the specifics of the application. "We file patent applications on a variety of ideas that our employees come up with," he said via email. " Some of those ideas later mature into real products or services some don't. Prospective product announcements should not necessarily be inferred from our patent applications."
The 2012 patent application offers details on more than a dozen separate metrics the company uses to rank news stories created by other Websites. How Google evaluates stories has been a point of contention with various media companies who have in the past said the company is infringing on their work. Many have also complained that Google can effectively turn on or off a spigot of visitors to a Web site by prominently displaying, or downplaying, a story. Google's decisions also affect what stories readers see, potentially shaping their view of news events.
Since it was launched in 2002, Google News has become the largest aggregator of news content on the Web. The site, which is completely computer-generated, collects and display headlines from thousands of news sources around the world.
The metrics cited in the patent application include: the number of articles produced by a news organization during a given time period; the average length of an article from a news source; and the importance of coverage from the news source.
Other metrics include a breaking news score, usage patterns, human opinion, circulation statistics and the size of the staff associated with a particular news operation.
Also factored in are the number of news bureaus a news source has, the number of original named entities used in stories, breadth of coverage, international diversity and even writing style.
The patent application provides some much needed visibility into how companies like Google select and rank online content, said Sree Sreenivasan, professor of professional practice at Columbia University's Journalism School and the university's first chief digital officer.
"In the world of technology, so much is opaque. It's nice to have some clarity around this stuff," Sreenivasan said. He noted that some of the metrics Google appears to be using to judge the quality of a news source are the same kind of metrics editors would use in deciding whether to trust a publication or not.
He pointed to metrics like staff size and audience diversity as examples. Even Google's use of story length is a good metric, Sreenivasan said. At first blush, it would appear that Google is emphasizing quantity over quality, he said. But the reality is that many high-quality media organizations now generate more content than they used to. So using story lengths and word counts is valid, he said.
"It reflects today's journalism reality," Sreenivasan said.
In an article from The Atlantic last September, Google News executives said the site "algorithmically" collects stories from more than 50,000 news sources and attracts more than 1 billion unique users each week.
Many in the media industry, especially in Europe, have rankled at what they view as Google's leeching away of readers and advertising dollars with its Google News site. But few have so far blocked their content from being displayed there, though Google offers a fairly straightforward way to do so.
Google itself has offered minimal insight about the algorithms it uses to discover and rank news stories. All the company will say publicly is that articles are selected and ranked based on metrics such as how often and on what sites a story appears; freshness of content; location; relevance; and diversity. The company has claimed that it constantly fine-tunes its news ranking to ensure high quality content is shown.
Last year's application appears to be the latest example of that refining process, offering a rare look at the some of the key ingredients Google considers. For instance:
- To determine the quality of a news source, Google could either look at the number of original ("non-duplicate") articles produced or count the number of original sentences produced by the source.
- To determine the importance of coverage, it can consider "story size scores" for all original articles produced by an organization over a week, a month or longer. "As an example, ...if D is an article about the crash of the Columbia Shuttle and there were 500 other distinct articles on the subject, then the story size would be 500."
- To calculate a "breaking news score" for a news operation, Google "may measure the ability of the news source to publish a story soon after an important event has occurred...."
- And to measure an operation's ability to produce high-quality, original work, the number of people mentioned in stories, especially if they're not widely cited elsewhere, could be used. "...This may be an indication that the news source is capable of original reporting."
Google also monitors links from search engines to individual articles. "Well-known sites, such as CNN, tend to be preferred to less popular sites, such as Unknown Town News, which users may avoid," the patent application said.
Jaikumar Vijayan covers data security and privacy issues, financial services security and e-voting for Computerworld. Follow Jaikumar on Twitter at @jaivijayan or subscribe to Jaikumar's RSS feed. His e-mail address is email@example.com.
Join the CIO Australia group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.
Cloud debate now about speed and sophistication
Yahoo Mail still down for some users, after an attempted fix
Queensland government to provide 200 services online by 2015
CIOs need to get their house in order, CFO panel says
Is Data Complexity Blinding Your IT Decision-Making?
5 Ways To Be More Productive At Work
Think back to the last time all your employees were in the office, at their desks, on the same day. It’s no surprise that you might struggle, between travel and off-site meetings, remote staff, flexible schedules and sick days. In today's competitive business climate, organisations need to maintain productivity and connectedness with their staff, despite not always being onsite. In this whitepaper, we look at five ways you can improve productivity, no matter where employees are.
Is your data centre growing too complex for your backup?
Backing up data today is growing more complex - and in an era of virtualisation, big data and cloud deployments, it can be difficult to maintain control over your data, resulting in loss and downtime. This hour-long webcast features expert commentary on navigating the complexity of backing up a heavily virtualised infrastructure; simplifying your backup software and hardware ecosystem; reducing the cost of backing up your organisation’s data, and modernising your backup infrastructure with integration. The presentations will conclude with an interactive Q&A session.
Deploying Customer Service in the Cloud
Customer engagement is rapidly shifting, and one of the biggest challenges companies are facing is figuring out how to serve consumers in this new digital environment. Implementing and managing new channels needs to be economically palatable, without any disruption to existing customer care practices. This business guide provides insight into how best to deploy customer service in the cloud, detailing the four phases of an evolved customer experience management strategy.