Method (140, 200) and apparatus (120, 270) for characterizing
digital content (124) using an
artificial neural network (ANN) engine (122, 274). Computer data sets (126, 128, 130, 132, 134, 160, 202, 232, 242, 302, 332) from a
library store (124) are processed to generate a corresponding sequence of multi-dimensional embedding vectors (162, 172, 182) in a latent space (170, 180). The embedding vectors are grouped into intervals or segments (166A, 168A, 228A) of the data sets based on movement
metrics (164, 166, 168, 228) associated with the embedding vectors. A representative vector, RV (174A, 184B, 210, 276) is selected for each group. Thereafter, in response to a query input (272), selected intervals among the various computer data sets are identified and output based on a
similarity measure (278) between the RVs and a search vector derived from the query input (150). Further embodiments provide a transformation model (322, 334) that transforms the embedding vectors and / or the RVs from a first latent space based on a first embedding model (304, 314, 332) to a different, second latent space based on a second embedding model (316, 338).