Skip to Main content Skip to Navigation

Extraction of Narrative Structure from TV Series

Aman Berhe 1 
1 TLP - Traitement du Langage Parlé
LISN - Laboratoire Interdisciplinaire des Sciences du Numérique, STL - Sciences et Technologies des Langues
Abstract : In the current explosions of multimedia content, there is a need to reorganize a large collection of multimedia documents. Narratives and their structure can help in this regard, particularly, in narratively long and continuous multimedia collections, such as TV series. Current TV series are composed of complex structures involving several intertwined narratives within the same episode and this continues until the last episode of the TV series. In this thesis, we focus on extracting and understanding narrative structure of TV series considering the fusion of multimodal features and narrative elements. Narratives in TV series come in small narrative units, scenes. Hence, working on scene level is the best way to extract and understand the overall narratives. In this thesis, we have proposed a novel way of extracting and understanding narrative structure via scenes linking. To do that, we have investigated scene segmentation which is based on features extracted from neural network models that take into account the visual and textual modalities of TV series. Then, a new way of scene-linking via fuzzy clustering is investigated, at different levels of granularity. The fuzzy clustering takes into consideration the narrative elements of scenes to create the links between them. Inter and intra episode links are created to capture the progression of a narrative throughout the TV series. Next, the most reportable scenes (MRS) are detected, to spotlight the turning points of the narrative structure on the linked scenes. Deep and complex neural network models are investigated for MRS detection by taking multimodal features of scenes. Finally, a visualization and evaluation tool is proposed to display extracted narratives and evaluate them according to a third-party, human intervention.
Complete list of metadata
Contributor : Aman Berhe Connect in order to contact the contributor
Submitted on : Friday, December 10, 2021 - 9:01:45 AM
Last modification on : Saturday, May 7, 2022 - 3:38:39 AM
Long-term archiving on: : Friday, March 11, 2022 - 6:19:46 PM


Files produced by the author(s)


  • HAL Id : tel-03474054, version 1


Aman Berhe. Extraction of Narrative Structure from TV Series. Linguistics. Université Paris-Saclay/Université Paris-Sud; LISN; CNRS, 2021. English. ⟨tel-03474054⟩



Record views


Files downloads