Yes, I call this method a “trainer”. However in other apps we have done we pick popular videos from several clusters based on similarity of usage. This reaults in a list of video (music in your case). Then allow the user to pick or like/dislike the music. After this training, which can go on as long as the user wants to do it, you have genre preferences (the genre of the music they picked) and some artist preferences (the artists of the music they picked) but also high quality music preferences to seed recommendations.
Picking genre, and artist alone, leads to, shall we say fuzzier knowledge of user preferences, by this I mean they don’t like everything from that artist, right? And remember that the typical behavior of a recommender is to not recommend something the user has converted on. In the case of video that’s probably ok since we don’t watch the same movies over and over (unless you're under 10 years old) but for music you would so turn off that feature. I’ll leave it to you to read the docs for config settings here: actionml.com/docs/ur BTW you need to record several events like genre-pref, artist-pref, as well as music-pref or better yet, play-50 if you have the data. The idea On Mar 16, 2017, at 5:56 AM, Masha Zaharchenko <[email protected]> wrote: Hi, everyone! I wonder if Universal Recommender is able to solve the following case. We ask a new user to choose genre/artist/writer/etc. she likes from a list or a bubble(e.g. ITunes) of options. This way we get "user-likes-genre/artist/writer" events. Also we have event history of purchases where the given entities(genre, artist,etc.) are in the properties of a product. Will we be able to recommend items to user immediately after this choice? E.g. a new user liked rock genre and got AC/DC recommendation right off the bat. Thanks, Maria
