Google To Predict Demographics Based on Video Watching Habits

Last Updated Aug 2, 2010 1:47 PM EDT

Couched as a way to help users find videos that would interest them, Google (GOOG) has worked on and filed a patent application for a predictive engine to determine demographics for people based on what videos they watch. The system would further allow Google to predict what demographic groups would be more likely to watch a new video, regardless of any headlines, descriptions, or other "meta" data that tries to describe the video.

A patent application originally filed in early 2009 and made public last week explains that the system would use a "training set of videos" to correlate video preferences to demographic information. The prediction model would let Google determine which demographic groups would statistically be likely to enjoy a new video that wasn't part of the training set.

Although described as a way to "suggest which videos would be of interest to a given user," it seems at least as likely that the intent is to help better direct advertising to accompany new videos as users post them on YouTube.

Instead of focusing on keywords and video descriptions, which can be missing or inaccurate, the Google approach would take "quantitative information on visual and/or audio features of the videos." The company would correlate that information to viewer demographics pulled from profiles -- presumably either ones users create or profiles that Google assembles based on the massive amounts of data it collects on consumers.

The training set of videos act as a calibration system. Google could how people react: which videos they pick, how long they watch the ones that they choose, and whether they comment or otherwise rate the videos. Then Google would extract the quantitative information and match it to demographic patterns most likely to respond. Finally, Google would examine videos recently uploaded to YouTube, extract the same quantitative information, and determine the most likely audience demographics. That would let the company in real time improve video recommendations for users and also better match videos, audiences, and advertising.

Additionally, by watching viewing patterns of consumers, Google could potentially use a type of triangulation to estimate in which demographic groups a person without a profile might fit. In theory, Google could also offer a service to video producers of recommending what visual and audio features to use to attract a given audience. That could be an interesting offering for television and movie producers, advertising agencies, computing gaming companies, and other industries that make heavy use of video and that are interested in specific user demographics.

As the application indicates, Google could not only look to its own store of videos on YouTube, but, in the web crawling of the search engine, analyze videos on other sites.


Image: user pirshulet, site standard license.
  • Erik Sherman On Twitter» On Facebook»

    Erik Sherman is a widely published writer and editor who also does select ghosting and corporate work. The views expressed in this column belong to Sherman and do not represent the views of CBS Interactive. Follow him on Twitter at @ErikSherman or on Facebook.