if there is a user X love actions movies so the algorithm will reccomend actions movies but if this user X strats to like new type of movies should i update this user profile and get a new feature vector for him or i should retrain the model ?
You should re-train the model.
ok but why retrain the model again the neural newtork has already learned to get user vector from user profile so if i update this user profile it should output new vector satisfy his new prefrences?
Retraining will let all of the users get better predictions using data from the new user.