Jinni claims to help you find movies, TV shows matching your taste watch online. I spent a few minutes training it and it’s not that great yet, but I do have eclectic tastes which might mess with its algorithms a bit and the system is obviously still growing.
What is interesting about it is its combination of expert opinions/classifications, machine intelligence (they talk of a movie genome that uses a rich ontology, akin to the music genome that led to Pandora) and collaborative filtering. It is clearly trying to marry the top down and bottom up in an interesting way. The model they seem to be using allows for the bottom-up to become more prominent as time goes by. I suspect that, as the user-base grows and the cold-start problem lessens, that this might turn out to be quite useful.
In its combination of sophisticated (and apparently recursive) algorithms and human input it is a fine example of a collective application. The use of two distinct strata of human input (the experts and the rest of us) gives an extra twist and a potentially richer dynamic than the usual fare.
Its use of an ontology offers benefits of parcellation as well as a richer set of ratings than the usual ‘this is good’ approach. In addition to the usual movie metadata, the main divisions are ‘experience’ and ‘story’, with each aspect subdivided into many other subtypes. The ‘experience’ aspect is particularly interesting, parallel in some respects to my own CoFIND system’s use of qualities, albeit in a more structured and less user-led form. The structure serves a purpose, though, allowing them to automate tagging once the system has been trained. If it works, this might help to overcome the problem of spiralling complexity and everlasting cold starts that have proved to be a stumbling block for CoFIND.
I look forward to seeing how this develops.
Created:Mon, 26 Jan 2009 18:10:57 GMT