Scalable Understanding of Multilingual Media Steve Renals - - PowerPoint PPT Presentation

▶

Jun 11, 2023 281 likes •430 views

Scalable Understanding of Multilingual Media Steve Renals University of Edinburgh Funded by the EU H2020 ICT Programme under Grant Agreement 688139 http://summa-project.eu Funded by the EU H2020 ICT Programme under Grant Agreement 688139

SLIDE 1

Funded by the EU H2020 ICT Programme under Grant Agreement 688139 http://summa-project.eu

Scalable Understanding of Multilingual Media

Steve Renals

University of Edinburgh

SLIDE 2

Funded by the EU H2020 ICT Programme under Grant Agreement 688139 http://summa-project.eu

SLIDE 3

http://summa-project.eu

SUMMA in a nutshell

Significantly improve media monitoring, by

the automatic

analysis of media streams across many

languages

aggregation and distillation of stream content
construction of knowledge bases from reported

facts

supply of media data visualisations at scale

SLIDE 4

http://summa-project.eu

BBC Monitoring

300 journalists each monitoring up to 4 TV channels several online text sources

30 languages – most important include Russian Arabic Farsi

SLIDE 5

http://summa-project.eu

Big Data

250 video channels
2.5Tb/day, 19Tb/week, 1Pb/year
BBC monitoring has access to
1,500 TV channels
1,350 radio sources
But… ~700 free-to-air Arabic satellite channels,

increases at ~100/year

Current monitoring processes are largely manual

and cannot keep up with the scale of the task

SLIDE 6

http://summa-project.eu

Use cases

1. External Media Monitoring
identify emerging trends
tracking people in the news
monitoring the evolution of storylines
2. Internal Media Montoring
manage multilingual content creation
efficient reuse of content across languages
3. Data Journalism
use SUMMA platform for data driven journalism

SLIDE 7

http://summa-project.eu

SUMMA Prototypes

Channel ID & native language Semantic Tag word cloud- size indicates current frequency across region/ group Segment Unique timestamp Player (Sd? HD?) Player controller - tag instances marked Tools - screen grab, snip video, save, attach Translated transcript “Now playing” text highlighted Tags shown underlined Add new tag - click pencil to ‘underline’, and enter text Segment machine analysis confidence (possibly better represented graphically?)

UI Concept 1

SLIDE 8

http://summa-project.eu

SUMMA Prototypes

SLIDE 9

http://summa-project.eu

SUMMA Prototypes

SLIDE 10

http://summa-project.eu

SUMMA Prototypes

SLIDE 11

http://summa-project.eu

Platform & Technologies

Speech recognition Machine translation Segmentation & clustering Ingest audio, video, text Identify entities & relations Summarisation & distillation Sentiment detection Visualisation & prototypes

SLIDE 12

http://summa-project.eu

Multilingual technologies

SLIDE 13

http://summa-project.eu

SUMMA system v0.1

SLIDE 14

Funded by the EU H2020 ICT Programme under Grant Agreement 688139 http://summa-project.eu