Taivastiuku
>I liked image x, what are some other images like x?
I created small service which answers this. It calculates image similarity by using tags and favourited_by_users data. Please try it out. Image id_number is used as query.
Example:
>>445463
http://derp.mlfw.info/for-image?id_number=445463
Username can be added to highlight already favourited images:
http://derp.mlfw.info/for-image?id_number=445463&username=Taivastiuku
The suggestions seem pretty decent. The real problem is updating the similarity database in a timely fashion. Currently I just load the whole derpibooru database through the API and reprocess all the data. Needless to say but this takes several hours and probably causes unnecessary server load on derpibooru (I do respect robots.txt and other instructions). Some kind of database diff could be nice or maybe I should just devise a scheme for determining which images have most probably changed and update those.
Another topic I’ve been studying are personal recommendations based on user fav history:
>Based on my history, what are some good images?
>Based on my history, what are some good tags or filters?
Currently the largest problem is that there seems to be no way to authenticate derpibooru user to a third party service. Giving out personalized results to anyone who knows a username doesn’t seem like a good idea.
Maybe the api key could be used? User could paste their key and I could call something like http://derpiboo.ru/whoami?key=123 to get the users name?
Best option would be oAuth2 or similar.
Option for user to export their downvotes would be nice too.
I created small service which answers this. It calculates image similarity by using tags and favourited_by_users data. Please try it out. Image id_number is used as query.
Example:
>>445463
http://derp.mlfw.info/for-image?id_number=445463
Username can be added to highlight already favourited images:
http://derp.mlfw.info/for-image?id_number=445463&username=Taivastiuku
The suggestions seem pretty decent. The real problem is updating the similarity database in a timely fashion. Currently I just load the whole derpibooru database through the API and reprocess all the data. Needless to say but this takes several hours and probably causes unnecessary server load on derpibooru (I do respect robots.txt and other instructions). Some kind of database diff could be nice or maybe I should just devise a scheme for determining which images have most probably changed and update those.
Another topic I’ve been studying are personal recommendations based on user fav history:
>Based on my history, what are some good images?
>Based on my history, what are some good tags or filters?
Currently the largest problem is that there seems to be no way to authenticate derpibooru user to a third party service. Giving out personalized results to anyone who knows a username doesn’t seem like a good idea.
Maybe the api key could be used? User could paste their key and I could call something like http://derpiboo.ru/whoami?key=123 to get the users name?
Best option would be oAuth2 or similar.
Option for user to export their downvotes would be nice too.