Neon Labs Inc. Software Predicts Your Pick To Click

By John D. Oravecz . This article originally appeared on Trib Live, October 17, 2013.

Editors know that images help draw consumers to news and social media websites.

Sophie Lebrecht, CEO and co-founder of Neon Labs Inc., a Carnegie Mellon University spinoff company, says using powerful images as thumbnails — the small photos that link to video or text — improves the frequency of clicks on them and leads to advertising revenue.

Neon developed technology that automatically selects the best frame in a video to become a thumbnail. That’s important, because online video viewership is booming.

Market researcher GfK said last month that more than half of those ages 13 to 54, the core TV viewing population in the United States, watches streaming video in a given week. GfK said 51 percent watch TV programs or movies using streaming video, up from 37 percent three years ago. Tablets and smartphones have aided in that growth.

“Images are a great way to draw people into the story, whether it’s a video or other content, and thumbnails will change the number of people that click to view that content — so the image you use for a thumbnail is incredibly powerful,” Lebrecht said.

Neon Labs is testing its software and intends to release a final version for sale this fall, said Lebrecht, who moved last year to Carnegie Mellon’s Silicon Valley campus in Moffett Field, Calif. Co-founder and senior technical adviser Michael Tarr, a professor of cognitive neuroscience and co-director of the Center for the Neural Basis of Cognition, works in Pittsburgh. The center is a joint program between Carnegie Mellon and the University of Pittsburgh.

To determine what people are likely to watch, researchers in Pittsburgh and at Brown University in Providence, R.I., studied how people process information — “how people see the world … how the brain generates a choice, that was our initial research question,” Lebrecht said. “Then we asked, can we build a computer model of the human brain that predicts how people will respond to images?”

A National Science Foundation grant enabled Lebrecht to participate in a program for entrepreneurs and venture capitalists, where she met investors. Last year, Neon raised $650,000 to continue its development.

“It’s exciting to have the opportunity to invest in a product that is grounded in hard science and can have such a tremendous impact on the media marketplace,” said John Burke, a founder of True Ventures in San Francisco, Neon’s lead investor.

Initial test results show that a thumbnail chosen by Neon significantly increased clicks, compared to one selected by a Web designer, Lebrecht said. Against a randomly generated thumbnail — the way most video platforms select thumbnails — Neon increased the clicks even more, Lebrecht said, although she declined to say how much.

Early on, YouTube and other websites focused on establishing a presence and building their capabilities, Lebrecht said. As video use grew, media publishers tried to maximize viewers.

“Each video has thousands of frames, and with video growing so rapidly, it will be hard for humans to process all of that,” she said.

Neon’s automated technique is one of the first to use cognitive and brain sciences. The underlying technology is co-owned by Carnegie Mellon and Brown, and licensed through CMU.

“Publishers will have the option to choose amongst Neon’s recommended thumbnails, or Neon will simply select one that is predicted to drive click-through rates,” Lebrecht said. “Seeing an image generates an emotional response that determines — click or no click.”

Tarr said algorithms used in Neon’s software are based on behavioral data obtained from surveys of hundreds of people and proprietary computer analyses of video streams.

“That’s our secret sauce, using machine learning techniques,” he said. “It’s how Google predicts what you most likely want to see.”

John D. Oravecz is a Trib Total Media staff writer. Reach him at 412-320-7882 or

Read more:
Follow us: @triblive on Twitter | triblive on Facebook