Guest comment: Moving recommendations beyond user profiling

With the mass exodus of customers from online profiling vendor Phorm, many people are questioning how relevant user profiling can be to the TV industry.  TV companies have been looking to various recommendations technologies as a potential boost to revenues over the last few years. The idea is that the more we know about a viewer, the better advertising and viewing recommendations can be targeted to specific users. This can be a simple case of targeting Desperate Housewives DVDs to people watching the show on TV or promoting a new US comedy to viewers of House.

Most of these technologies are derived from call centre or internet applications. A typical example is a mobile phone call centre, where the users profile is analysed when they call up, and the most relevant tariff and call plan is offered based on the user's profile. High volume callers are offered different bundles to users who use SMS.

The most well known internet recommendations are Amazon.com's "People who like this also like this" service. In this case, Amazon compares a users shopping profile with that of other visitors and makes recommendations based on their shopping habits.

Deploying such a service in the TV environment typically involves implementing a recommendations engine into the infrastructure of the TV provider, where the engine can collect information on what users are watching to analyse their viewing habits and subsequently make recommendations.

On the surface this sounds fine, but the critical difference between TV and online services like Amazon is that TV viewers are being tracked on every show they watch - regardless of who might be watching at the time, and regardless of what they are watching.

Most viewers have guilty pleasures in their viewing, be it shows they watched as a child or the occasional adult movie, and they typically don't want these to be used for their household recommendations. The basic problem here is that most recommendations track what people watch and make recommendations on this basis.

Phorm's business case has collapsed because of user uproar during trials: it would seem that the people of the UK do not like having their behaviour stored and profiled. The privacy concerns are substantial and many saw the UK as a more liberal test market than other parts of Europe so a widespread rollout now seems unlikely.

This leaves TV companies stuck between a rock and a hard place: they need recommendations technology to boost multichannel viewing and VoD revenues but now have substantial risks associated with user profiling.

The only way through this is to develop recommendations methods that do not rely on user profiling. An integrated approach that can provide personalised recommendations to each user without the need for user profiling.

By analysing people's viewing habits solely with the intention of building a map of how different TV shows relate to each other we can get the data we need. If a lot of people who watch Desperate Housewives also watch Sex in the City then this gives us a strong correlation between those two shows.

We've been developing this for a while and it works. 

Where this approach is different to Phorm is that when it comes to making a recommendation, we don't take into account any viewing history or profile information, making the recommendation solely based on the show the user is currently watching, or any specific favourite shows they have set as preferences.

This approach gives the broadcaster all of the benefits of targeted TV recommendations without the privacy concerns of profiling. The point is that the much needed recommendations engine for online viewing can work. Indeed, it's a powerful tool. However, it can only work effectively if the viewing public buy into it and feel comfortable with the methodology.

By Tom Weiss

CEO      

TV Genius

www.tvgenius.net

********************************

Add Netimperative to Twitter