In Figure eight two movies with the highest average illumination and one film with the lowest illumination are offered. For the supervised models (TAM and GraphTP), the common D is normally decrease, which signifies that they're able to adapt to the positional bias and choose occasions from all components of a movie. 2020); TAM is sequence-based mostly supervised mannequin which employs a sliding context window and computes the similarity between sequential contexts.