Posts
From Niemanlab.org:
"Blogs and newspaper sites enjoy a built-in advantage when it comes to search-engine optimization. They deal in words. But a whole universe of audio and video content is practically invisible to Google.
Say I want to do research on Osama bin Laden. A web search would return news articles about his assassination, a flurry of tweets, the Wikipedia page, Michael Scheuer’s biography, and an old Frontline documentary, “Hunting Bin Laden.” I might then take my search to Lexis Nexis and academic journals. But I would never find, for example, Frontline’s recent reporting on the Egyptian revolution, where bin Laden makes an appearance, or any number of other video stories in which the name is mentioned.
While video and audio transcripts are rich for Google mining, they’re also time-consuming and expensive. PBS is out to fix that by building a better search engine. The network has transcribed and tagged, automatically, more than 2,000 hours of video using software called MediaCloud.
“Video is now more Google-friendly,” said Jon Brendsel, the network’s vice president of product development. Normally, automatic transcription is laughably bad — Google Voice users know this — but Brendsel is satisfied with the results of PBS’ transcription efforts. He said the accuracy rate is about 80 to 90 percent. That’s “much better than the quality that I normally attribute to closed captioning,” he said. The software can get away with mistakes because the transcripts are being read by computers, not people. (For a hefty fee, the content-optimization platform RAMP will put its humans to work to review and refine the auto-generated transcripts.)..."
No comments:
Post a Comment