Suggestion engines are the holy grail of interactivity – they provide the “smarts” in smart interactions. They exist in various forms today, but not in the form necessary to drive the evolution of the video market, either for convergence of divergence of TV and web video. To be competitive web video has to be ultra simplistic and allow for the same type of passive viewing as television where people “just turn it on and watch.” The ideal system is a combination between what Tivo and Amazon already developed and the application of tagging in blogs or YouTube. For digital TV to evolve, it’s necessary to have more advanced features to suggest viewing in order to flesh-out features like parental controls and on-demand viewing.
Search Engines from then to now …
The basic idea of a search engine with intelligent suggestion is not only just a good idea, it’s arguably been vital to the advancement of the web for just about everything. Suggestions engines are familiar to the web; the most famous suggestion still in use is Amazon’s. There were a slew of super hot startups in the space in the 90′s that came up empty – none of them had any genuinely original technology. Worse it was one of those spaces that became cluttered with confusing claims that further frustrated the advancement of the right tools.
The many efforts to master suggestions have done little more than create various profiling systems coordinated with search terms to help refine search processes. The main problem is that the web cookie metaphor doesn’t define who is searching; it only defines the specific web browser on a computer. The cookie uniformly stores all the information on historical activity for that browser with the same relevance; therefore the engine collects an aggregate of searches jumbled together. There is no way to discern the relevance or the relationship of unique search queries.
In my case I am often a Dad looking for family things, but often I am wannabe college boy again looking for trouble – two totally different personas. To get around this, the search tool would need to know the searcher’s identity, and that starts a whole new set of requirements and restrictions on UI behavior.
Case Studies
My strongest memory of how a suggestion engine can be a disaster is my own experience around the time my first child was born. I was (and am still) convinced Lamaze was a scam, a way to get people to pay for classes they don’t really need. So I set out to find a video Lamaze course so we could check out Lamaze for a lot less than paying for classes in Manhattan. I also at the same time was collecting hard to find music that I figured would go out of print and disappear (I have never been a fan of filesharing; I wanted the CDs). The result has been that for years Amazon has been trying to sell me anything and everything related to Paul Anka. I suppose in the over simplified perspective Paul Anka is a musician who wrote a song about having a baby, but in my mind the association was downright comical – especially considering the genre’s of music I was searching. That’s the danger; the associations made by computers are not often reasonable, reliable or realistic.
The most legendary example of suggestion engines gone awry is the old riff “Tivo thinks I’m gay.” In this case the suggestion engine is “too fuzzy” in its associations. Here, the suggestion tool works with very general descriptive tags about something and associates them too stringently to the searcher. The key design issue here is that the searcher doesn’t know what tags are being associated with their choices, or most important: why that’s the description. For instance the “Tivo thinks I’m gay” pattern probably arises from the consumer picking subjects that have common tags, like “romantic comedy” for watching reruns of “Friends.” It’s not actually a legitimate assumption that the person viewing choices are specifically because of an interest in romantic comedies. There interest could be uniquely “comedy,” or any number of reasons completely not associated with the fact that what they are viewing what someone else considers a romantic comedy.
The must-have feature: profiling
It’s easy to see how a collection of unusual circumstance could radically foil any effort to make sensible associations. The conclusion is that for suggestion engines to work, there has to be some form of profile management for unique users; it becomes unavoidable.
For the web, I would like to see a suggestion engine with a short memory that I can turn off or on. This would not require me to establish my identity to the search engine, what is established is the profile for a search session. This would allow me to associate a sequence or set of searches. Ideally, I could have some sort of universally portable “light profile” that allowed me to keep my search interests and biases, and even gave me the option to edit them. This way I could mix aspects of searches as I refine them, or call up old searches, save them and continue to refine them. This is a missed opportunity for search providers.
Why this hasn’t happened
Search providers have all launched their ships into being “portals.” In their vision of a portal they are all things to all people and diving into mail, calendars and business collaboration, each asks the consumer to walk deep into their web, changing the brand message from being the leaders in search to being everything web. Their approach is surely more lucrative, but the value of offering superior search is not served.
Enlightened self-interest … of course
My professional interest is as someone building video blogs. I’d like the see the creation of a system that allows the searcher / viewer to view a sequence of short videos – going from the first video they select and seamlessly flow to the next video in a logical sequence. There are a few sites that already do this. At the end of each video there is a delay and the next logical next video is queued. YouTube sorta does it, but its more random than programmatic. The logic of ‘what’s next’ is based on criteria considering other viewer’s previous viewing and search criteria. For instance the first video might introduce a concept, and the second video would queue to elaborate on the next logical question that most people would ask, and then play that after a few seconds. During that short intermission the user could also presented a handful of options of alternative things to view, like follow up videos that may answer their unique or different questions about the first. This could be a very potent tool for corporate sales and dramatically advance video … and yes, I have a developer working on it.
Some people call this non-linear entertainment, but I don’t think the real value is in entertainment; its for business use. The idea is that the viewer has the control to change a storyline or the sequence of viewing. This would be a game changer for video blogs and YouTube. YouTube would be hard pressed to overlay such a custom system onto what it currently operates – its tags are too general. This sets up my upcoming post on the longtail and the future of web-video.
Conclusion
The web is all about choice and knowing how to navigate. It’s too easy to make the wrong choice – with TV it’s just too simple to hit the channel button. TV has too much filler and with so many competing entertainment options for consumers; it needs to modernize its offering. The suggestion engine feature must be evolved is to help TV survive as well as create web-video mass adoption – converge or diverge doesn’t matter both media; need it. As Tivo proved, solving the profiling issue for a DVR is even something people would pay for – it’s my opinion all DVR’s need something like this. With the web … this isn’t that hard to implement technically; it’s the content that’s the hurdle.
