Recommendation algorithm/engine for self hosted media server. If someone knows where I can get a large amount of watch data for a large amount of people I could write such a thing.
I have 2 years tautulli logs for a plex server of 10 users. Only has 3-4 users regularly watch stuff and 90% is my kids watching that same thing over and over, hahhaha. Maybe one day I’ll find good use of the data.
I would probably try implement Netflix’s algorithm which is essentially finding what u haven’t seen that has been seen by people who watch things similar to what u watch. Essentially just heigh dimensional embedding and some clustering. So u would want at least 1000 (absolute bare minimum) users.
Data is only really usefull in the context of all other users. So 3-4 users of data on their own don’t really give any useful insights.
Totally agree my data has no value here :) I have considered charting genre viewing preferences and how it’s changed over the years but that’s years away.
Slightly related, I have a service that’s been tracking my music history for about 10 years. It’s fun to see how my tastes have changed!
The place u might find some interesting insights are the timestamps of when things get watched. If u got a health tracker I recon u might find interesting correlations between that and watch data.
How will u classify a taste? U can always eyeball it but perhaps embedding it within a vector dB might work. If u also dump in a large catalogue of just general population music sample u can get a full genre landscape and see how u fit into it.
Ohh that’s really smart! I’ve been tracking my workouts for almost 2 years. I really like the idea classifying my taste in music. Will definitely my it on my list of projects!
Recommendation algorithm/engine for self hosted media server. If someone knows where I can get a large amount of watch data for a large amount of people I could write such a thing.
Trakt.tv has a lot of data, and they provide recommemdations
I have 2 years tautulli logs for a plex server of 10 users. Only has 3-4 users regularly watch stuff and 90% is my kids watching that same thing over and over, hahhaha. Maybe one day I’ll find good use of the data.
I would probably try implement Netflix’s algorithm which is essentially finding what u haven’t seen that has been seen by people who watch things similar to what u watch. Essentially just heigh dimensional embedding and some clustering. So u would want at least 1000 (absolute bare minimum) users.
Data is only really usefull in the context of all other users. So 3-4 users of data on their own don’t really give any useful insights.
Totally agree my data has no value here :) I have considered charting genre viewing preferences and how it’s changed over the years but that’s years away.
Slightly related, I have a service that’s been tracking my music history for about 10 years. It’s fun to see how my tastes have changed!
The place u might find some interesting insights are the timestamps of when things get watched. If u got a health tracker I recon u might find interesting correlations between that and watch data.
How will u classify a taste? U can always eyeball it but perhaps embedding it within a vector dB might work. If u also dump in a large catalogue of just general population music sample u can get a full genre landscape and see how u fit into it.
Ohh that’s really smart! I’ve been tracking my workouts for almost 2 years. I really like the idea classifying my taste in music. Will definitely my it on my list of projects!
If u do pls publish the code under a foss licence and send me a link.