Channel-surfers take heart. Now there is a Web site that will scour the TV listings, pick out a collection of programmes it knows you will enjoy and then deliver them in a personalised television guide.
The PRISM Laboratory in the Department of Computer Science at UCD has developed a programme with sufficient artificial intelligence to make your television choices for you. It builds up a profile of your favourite viewing, learns the finer points of your taste as you review its selections and can identify wholly new programmes that it `thinks' you might like.
"The more you use the system the better it knows you and the higher quality the recommendations," explained Dr Barry Smyth of the PRISM Lab. And, the better the recommendations, the more likely the user is to return to the site and use the service, known as PTV, Personalised TV Service.
The Lab involves two UCD staff and about 20 students. It is a research group within the Department "focusing on artificial intelligence and multimedia systems," explained Dr Smyth. "We generate ideas for areas we would like to work in," and then finance is sought to enable the research to get underway.
"This particular project began three years ago. At that time I was particularly interested in the Web and the Internet." He saw that the Web was beginning to suffer from a problem now familiar to any user - information overload. "The Web was becoming too popular for its own good. It was becoming more and more difficult to find what you wanted."
The answer, he believed, was to introduce a level of information preference, the ability to personalise the information retrieval activity. "We want to filter out the information that is not relevant and light up the information that is," he said. "What I am saying is that this extra layer of personalisation is going to come from artificial intelligence."
The group began looking for areas where the personalisation of information searches could be prototyped. "One of the ideas that jumped out was to personalise TV listings," a service that is particularly relevant given the imminent introduction of digital TV and the great variety that it promises.
The result of the team's efforts is now resident at www.cs.ucd.ie/ptv and using it couldn't be easier. The site walks you through a registration procedure which helps it to know what you want to watch. It asks what channels you watch, when you are available to sit in front of the tube and the kinds of programmes you particularly want to see - or miss.
You are given a username and then, each time you login, the site will give you a personalised viewing guide. There are Info icons which provide details about the shows on offer and PTV will also offer recommendations for you to try. When it does this it will ask for feedback in the form of your own grading of the recommendation. The system uses your response to adjust its profile for you and it should make increasingly better recommendations over time, unless of course you are a completely fickle viewer.
The prototype was designed, built and moved onto the Net about a year ago and since then has attracted about 1,000 registered users who hit the site a total of about 100 times a day. There had been no plans to keep it running, Dr Smyth said, but there was a great deal of positive feedback. "Then we realised that it was really useful and that people liked to use it."
A decision was taken to put it on the Net in an official way. The team set about improving the interface and the current version of PTV was installed about a month ago.
There is a substantial level of artificial intelligence to drive the profiling, he said. "The system can learn to adapt to the user and builds a sophisticated user profile," Dr Smyth said. It was based on two types of reasoning, allowing it to offer "content-based" recommendations and "collaborative" recommendations.
It looks at the TV programming on offer and, using the user's preferences, can look for matches, for example Friends or Star Trek or news programmes. It isn't like an Internet search however and a letter-for-letter match was not essential. "It is interested in similarity, not exact matches." Thus Friends could also mean comedy.
When the system looks at collaborative choices, it ignores content and selects instead on the basis of your preferences as a viewer and similar preferences exhibited by other users. If you match another user in several ways, the system might make a recommendation known to appeal to that user, anticipating that you might also enjoy it.
PTV is also highly sophisticated in the way it sources information about viewing choices. It automatically sends "web agents" to interrogate known TV listing websites, for example RTE's site, and copies across the viewing information like a human user would. It can also dip into the Internet Movie Database (us.imdb.com) for more information about films on offer as it attempts to decide whether a given user might watch it.
The artificial intelligence developed to support PTV had numerous other possible applications, Dr Smyth said. The research group has already entered discussions with a number of potential users for things such as personalised recruitment, distributing Web advertising to users on the basis of established interests or preferences and incorporating the artificial intelligence aspects into other listings-type sites.
The object in all cases would be to reduce the information overload and make it easier for people to make best use of the information resources on the Web.
Dick Ahlstrom, Science Editor: dahlstrom@irish-times.ie