PORTALE DELLA DIDATTICA

Ricerca CERCA
  KEYWORD

Blobs Remix: Narrative Blends of Archives and Social Media

azienda Tesi esterna in azienda    


Parole chiave AI, CINEMA, ARCHIVE, SOCIAL MEDIA

Riferimenti TATIANA MAZALI

Riferimenti esterni Lorenzo Canale (Centro RIcerche Rai)

Tipo tesi SPERIMENTALE - PROGETTAZIONE

Descrizione “Blob” is a television program that creatively edits short fragments of other TV shows to generate new
meanings and messages. The proposed project aims to automatically create new video clips that apply
“Blob”’s editing style integrating fragments from social media videos and archival footage. To reach
this goal a neural network will be trained to recognize instances where two fragments emulate “Blob”’s
distinctive editing style. Finally, these newly generated clips will then be published on social media
platforms to gauge their impact and reception among audiences. This approach not only showcases the
innovative capabilities of AI-driven content creation but also provides insights into how Blob’s style can
resonate with contemporary media audiences on digital platforms.
WORK PLAN
1. Creation of Training Dataset and Video Collections (2 months):
(a) Dataset Creation for Neural Network Training (1 month and an half): The first task involves
creating a dataset (D) to train the neural network. This dataset should include numerous
examples of cuts from the “Blob” program as well as fictitious cuts, extracted from other
contexts or invented, allowing the network to differentiate between good (“Blob” similar) and
bad (“Blob” different) cuts . Given that the entire (“Blob” similar) program is not annotated
with cut timestamps, this may require a hybrid effort using both automatic scene change
detection techniques and manual work.
(b) Creation of Video Collections (half a month): The second task involves creating two video
collections: one from archival sources (Ca) and another from social media sources (Cs). These
collections will serve as testing grounds for evaluating the algorithm’s performance.
2. Neural Network Training (1 month): This month will be dedicated to training various neural
network configurations to achieve the best results in discerning between the two types of examples
in the dataset D.
3. Automatic Clip Creation and Social Impact Assessment (2 months): Automatic clip creation involves applying the trained model to automatically create clips from both Ca and Cs and
selecting the clips that appear most relevant while keeping track of the criteria used to select them.
For Social Impact Assessment, a TikTok channel will be created, as it is a social platform where it
is easier to reach more people in a short amount of time. Metrics such as views, likes, shares, and
comments will be evaluated. These two phases are interconnected because the selection of clips to
post can vary based on the results achieved with the audience.

Vedi anche  call_for_projects_fiatifta_centro ricerche rai.pdf 


Scadenza validita proposta 05/04/2025      PROPONI LA TUA CANDIDATURA