Kudos for publishing your approach to recommendations!
Relying on users to explicitly mark objects is not the best approach, though. Each user have their own mind about what the best marks/weights would be, hence the input would be inconsistent and the recommendation results would suffer. Not to mention that manual marking is tedious, and as you pointed out - it is hard to apply to old content.
I would suggest using a system that can extract the marks automatically. Couple of years ago I was involved in a project that had such a component (NeuralBrother's Neugs), but there are OS projects that provide similar functionality as well.
Relying on users to explicitly mark objects is not the best approach, though. Each user have their own mind about what the best marks/weights would be, hence the input would be inconsistent and the recommendation results would suffer. Not to mention that manual marking is tedious, and as you pointed out - it is hard to apply to old content.
I would suggest using a system that can extract the marks automatically. Couple of years ago I was involved in a project that had such a component (NeuralBrother's Neugs), but there are OS projects that provide similar functionality as well.