3 ms·
Human powered in the sense that the expert system we developed uses those human likes, views, star ratings and implements recommendation features to facilitate
by fjabre 7y ago
Human powered in the sense that the expert system we developed uses those human likes, views, star ratings and implements recommendation features to facilitate humans recommending titles to other humans in a very straightforward way.
But there is no ML compiling a profile of every genre/title you've ever expressed interest in and then using that profile to guess what you want to see next like Netflix et al.
- anjc 7y agoAt some point you presumably have to assess user-similarity in order to let similar users recommend items to each other? How's this different from e.g. user-user CF in terms of compiling everyone's info into a single model? I'm not criticising but rather just curious
- fjabre 7y agoI haven't found enough of a need for it at this point to make the application more useful as it stands without making great effort in training the necessary models to assess user-similarity for example. It is certainly faster to build an expert system that makes a lot of assumptions and iterate from there. There are plenty of use cases for ML and I hope to integrate it more into app some day to help automate repetitive tasks.