What is the solution to the problem of recommender systems?

Question

data mining

Task : A user downloads photos from a site. The user has some properties (eg gender, country, etc.). Each photo also has certain properties (for example, the number of views, downloads, etc.). A new user arrives. What photos do you recommend?
Given : A matrix of the following form:
User i{Vector of user properties i} {Vector of downloaded user photos i{for each downloaded photo, the vector of its properties}}
Required : for each user described by the corresponding matrix vector, obtain a recommendation rating vector, where the highest values ​​will be the most appropriate Photo.
At the moment, various solutions existing on the Internet have been tried, including the librec library (it also includes the well-known Slope One). However! They do not allow taking into account the properties of the user and the photo, that is, the input data contains only the User / Item matrix.
There was also an attempt to apply the Factorization Matrix algorithm, promoted by mail.ru, using the LibFM library, but in fact it turned out that it was not intended at all for the purposes of the recommendation, but as a classification / regression, since the format of the input data (as in the libsvm library) is : label (class) {idi}:{vali} {idi1}:{vali1}...
Question :
1. Is it possible (if so, how) to use libFM for recommended tasks. systems?
2. What solutions exist that allow using not only the user / item matrix, but also the properties of users and objects in the input data?