When a user enters her Netflix account, recommendations such as ‘Top Picks For ___’ and ‘Because You Watched ___’ are all over the place, and often spot on, based on certain algorithms.
This Netflix patent US8887095B2 describes a method for recommending shows based on collecting information during your current Netflix session, from an ‘endpoint device’ which can be computer, mobile device, or set top box.
The collected information from the ‘endpoint device’ can include location information, time of day, user selections and button presses in the current Netflix session, button press timing and pressure, and audio and video information collected through microphones of the endpoint device.
The collected information is then processed to infer things about the current user, such as the user’s identity, age, emotional state, gender, etc. For example, time of day may indicated that a child is using the endpoint device, or the pressure with which a button is pressed may indicate gender of the user.
The information is analyzed to match patterns of the collected information and user behavior in the current session with other users’ behavior. In other words, depending on the collected information, the algorithm can find other users with similar collected information and make recommendations to the current user based on what the other users with similar collected information watch. The process of finding patterns in the current user’s collected information, learning patterns in other users, and matching the pattern of the current user with previously learned patterns of other users, can be performed with machine learning (e.g., neural networks).