VIDEO CONTENT ANALYSIS FOR AUTOMATIC DEMOGRAPHICS RECOGNITION OF USERS AND VIDEOS

摘要

A method comprises: accessing a plurality of videos; retrieving a video access log; differentiating a viewers from the video access log; extracting demographic information from one or more viewers associated with the viewer that accessed the video; determining whether the demographic information includes demographic attributes of interest; generating and storing a multi-attribute demographic distribution for the viewer; generating a feature vectors for each of the videos based on video content of frames; associating each feature vector with a visual object within a frame within each video; generating and storing a subset of representative feature vectors by performing dimensionality reduction; in response to storing the subset of feature vectors and the multi-attribute demographic distribution for each video, generating a predicted distribution; and determining, for the viewer having extracted demographic information that does not include the demographic attributes of interest, predicted demographic attribute values using the predicted distribution.