Consumers all around the world used devices like TiVo to record television programs, as well as to rewind and fast-forward live television shows as they are being viewed. In addition to the ability to skim through programs, a TiVo unit can also provide “You Might Like This Program” suggestions to the viewer. These suggestions are based on a variety of factors that make up a viewer’s profile, as well as previous ratings by the viewer and by similar viewers.
Data Mining to Create a Personal Profile
TiVo uses data mining to create the profile of each viewer. Viewing history and ratings are both used to determine the chances of a viewer liking a new program. TiVo uses a rating system based on a thumbs-up or thumbs-down evaluation by the viewer. You can rate shows as you watch them, while you record them, or when you see the program in the guide. Providing ratings to TiVo allows the system to compile more information about your likes and dislikes, and to provide a better selection of other programs that you might like.
Simply basing recommendations on the ratings of a viewer can lead to discrepancies, however, according to Wijnand van Stam and Kamal Ali, authors of TiVo: Making Show Recommendations Using a Distributed Collaborative Filtering Architecture. In this 2004 article, presented at the 2004 International Conference on Knowledge Discovery and Data Mining, (pages 394-401) it is noted that TiVo also uses the history of shows a viewer has recorded to evaluate the viewer’s preferences. This is due to the fact that viewers apparently rate shows based on perception, rather than reality.
“Viewers have been known to rate shows that they would like to be known as liking but that they don’t actually watch. Some viewers may give thumbs up to high-brow shows but may actually schedule quite a different class of show for recording.”
What does this mean? People rate some shows highly, but never actually watch them. This would throw off your suggestion results, if TiVo didn’t take it into account in the sorting algorithm.
TiVo Recommendations Can Come From Other Viewer Ratings
In order to recommend programs, or arrive at a suggested rating for a program that has not yet been rated by the viewer, TiVo can also use information gathered from other viewers. By comparing the ratings for other shows given by the active viewer to ratings given by other viewers, it is possible to find viewers with similar tastes. TiVo then extrapolates a suggested rating for the new show, based on the ratings of those similar viewers.
Artificial Intelligence Improves the Viewing Experience
Using artificial intelligence tools, such as data mining, to improve the experience of consumers is an idea that is also being used in other industries with a great deal of success. Google uses data mining to create a customized search experience, for example. Netflix, an online video rental company, also provides ratings and suggestions similar to the TiVo model to customers. By using these tools to increase customer satisfaction, companies can increase their chances of retaining customers, as well as improving competitiveness in their chosen industry.
Bozdogan, H.; Statistical Data Mining and Knowledge Discovery. Chapman and Hall. (2003).
Witten, I, Frank, E. (2005). Data mining: practical machine learning tools and techniques. San Francisco, CA: Elsevier, Inc.
Decoding Science. One article at a time.