"I want this to be a force for a democracy. I want us to be an honest broker, and I want newspapers featured on our site to get traffic from us. There's never been a more controversial time on the planet. I think it's great to be a news source at this point because there's so much hunger for news. You see a lot more diversity in the news coverage on our site than on others . . . Even if we did want to bias it, fundamentally we are committed so strongly to objectivity we couldn't possibly do it. I think no matter what political association you belong to, it's valuable to see what the other side is saying."
Kirshna Bharat, in a 2003 interview with PaidContent.org's Stacy Kramer (link below).
Photo Linked From: http://research.google.com/people/krishna/krishna.jpg
Ever wonder how Google picks the news? Here's a clue: It all starts with people -- Krishna Bharat in particular.
Google has been relatively secretive about the way its general search service finds and ranks web information resources. But in an effort to reach out to news pubishers, the company has just produced a little blue pamphlet, "Google News," which answers the question: "How does Google news work?" In addition, Josh Cohen, business product manager for Google News, based in New York, was on a panel at the Online News Association's annual meeting in Toronto on Oct. 19.
"The headlines on the Google News homepage are selected entirely by a computer formula, based on many factors including how often and on what news sites a story appears elsewhere on the web," says text in the Google News blue booklet made available at the ONA meeting. "Basically, we look at the number of original articles being produced and published by editors in order to determine the size of a story, which we also weigh based on how recent it is."
For example, the booklet explains, Google News might register in a two-minute time period a "cluster" of two stories about a sumo match in Japan, and a cluster of 200 stories about U.N. talks with Burma. The human-written algorithm detects that the latter is a bigger story and gives that cluster priority in rankings.
"In the ballpark" of 50 people work for Google News, Cohen told the ONA audience. His discourse on the inner workings of the the news search was prompted by a question from fellow panelist Meredith Artley, executive editor of latimes.com. "How does the Google algorithm take in reputation and trust?" she asked.
Algorithm is a term used by software engineers to describe a series of instructions, or code, which set up the way a computer handles routine or repetitive processes, often in the form of a series of nested "if/then" statements which attempt to chart a path for the processor to take under all forseeable circumstances. Once complete, the algorithms don't require human intervention to operate. But the logic of the "if/then" processing is supplied initially by the people who write the lines of code. So while computers direct the ongoing news search at Google, the initial instructions are supplied by people.
There are two parts to the ranking process, Cohen said. First, the human-written algorithms compare the number of instances across the web that a particular news subject is appearing over a given period of time. That process, said Cohen, includes "who's publishing it, how often, what original content is being produced on that, and where is it on the website -- is it above the photo next to the big photo on the page?" Thus at this stage, Google news essentially factors in the news judgement of editors who have decided on story placement and headline and photo size on a web page.
"That is the cluster part of it and then there is the ranking in that cluster of an original source," said Cohen. "What we are trying to get is local sources, trusted sources."
As a story's ages across the web, another key factor figures in how much prominence it is given by Google News, Cohen said -- how frequently it gets "clicked" by Google users. A story which might have begin a few lines down in a subject ranking will start to move up if users ignores its ranking and click on it more frequently than the stories above it. Or a particular news source may be interpreted to be more trusted by users who favor it in their clicking, causing it to rise up up in rankings on that basis.
"If a user bypasses that first source and goes down to the third source, that tells you something about the user's trust in a given story," says Cohen. "The ranking is variable on a story by story basis."
Another factor is supplied in the algorithm-writing process, said Cohen -- a human judgement about the relevants of a source to a particular topic. For example, he said, Google news algorithm writers might have concluded that the Los Angeles Times should be ranked as a more trusted source for stories about the entertainment industry, because Hollywood is in the paper's backyard and it is a beat for several reporters. That fits a Google News policy of favoring "local" sources.
Do algorithm-writing instructions constitute a form of editing? And who are the humans who write or advise the writing of Google news-search algorithms? No one asked Cohen. But the blue booklet explains their origin.
"Google News was created by one of our engineers, Krishna Bharat, in the wake of Sept. 11," the booklet says. "Krishna's vision held true to the basic tenets of journalism: inform people about what's going on in the world around them, open their minds to new voices, and encourage them to look at events from different points of view."
Cohen himself is a former sales executive for Reuters, the British-based business and general news purveyors. Prior to that he was with the SmartMoney.com joint venture of Dow Jones & Co. and the Hearst Corp. He holds degrees from the University of Michigan and the Columbia University Graduate School of Business.