The present disclosure provides methods, computer-readable media, and devices for making content recommendations in a manner that protects private user data. In one example, a method includes collecting content consumption history data for a specific user of a telecommunications network, inserting random noise into the content consumption history data to produce anonymized data, sending the anonymized data over a network to a centralized device, receiving correlation data from the centralized device, wherein the correlation data indicates correlations among a plurality of items of content that is available for consumption via the telecommunications network, and predicting a likelihood that the specific user would be interested in a particular item of the plurality of items of content, wherein the predicting is based on the content consumption history data and on the correlation data. The steps may be carried out by a processor of a user endpoint device.