Video content modeling with latent semantic analysis