Open Password – Wednesday July 14, 2021
#947
vfm spring conference – Technical innovations – Automatic transcription of videos – Television 3.0 – Automated sentiment analysis – AI-generated metadata – Stefan Hauff-Hartig – Development of audio and video files – Fraunhofer Audio Mining – Christoph Schmidt – Fraunhofer Institute for Intelligent Analysis and information systems – ARD-Tagesschau – Live recognizer – Multilinguality – Deutsche Welle – Jakob Rosinski – Watson – IBM – Automated sports highlights – Automation of camera settings – Re-broadcasting – Media archivists – Joanna Bars – German broadcasting archive – Nikolaus Korfhage – Philipps University Marburg – Kader Pustu-Iren – TIB Hannover – VIVA – Face Recognition – Convolutional Network – Fesad – Dirk Maroni – Westdeutscher Rundfunk – Data Analytics – Content Analytics – Robot Journalism – WDR KIMD – Experience Evaluation – ARD Crossmedia Search – Future of Information Science – Practicality – LexisNexis Risk Solutions – True Cost of Financial Crime Compliance Global Report – US – Germany – Customer Risk Profiling – Sanctions Screening – Regulatory Reporting – Identifying Politically Exposed Persons – KYC for Account Onboarding and Alerts Resolution – Pandemic Impact – Technology Investments – BIIA – dpa – Fact checking – Fake News – Disinformation – Google News Initiative – Journalism – Alexandra Borchardt – Baybars Örsek
Cover story
Session: Technical innovations: Automatic transcription of videos – Television 3.0: Automated sentiment analysis and compilation of short videos with a high level of excitement – AI-generated metadata: From technology observation
to productive use – By Stefan Hauff-Hartig
II.
LexisNexis Risk Solution: Global Spend on Financial Crime Compliance Reaches US$214 Billion
III.
dpa: Fact checking: Against fake news and disinformation
vfm spring conference 2021
session: Technical innovations
Automatic transcription of videos
Television 3.0: Automated sentiment analysis
and compilation of short videos with high levels of excitement
AI-generated metadata:
From technology observation to productive use
By Stefan Hauff-Hartig
Stefan Hauff-Hartig
The third session, moderated by Michael Vielhaber from Austrian Broadcasting Corporation, introduced participants to pioneering tools and concepts for AI-supported development of audio and video files. All four technologies presented are already proving themselves in their practical application environments.
__________________________________________________________________________________
Fraunhofer Audio Mining
___________________________________________________________________________
First, Christoph Schmidt from the Fraunhofer Institute for Intelligent Analysis and Information Systems presented the “Fraunhofer Audio Mining System,” which is used to index media data, make it searchable and enable faster navigation. Using a sequence from an ARD Tagesschau, he showed how the system analyzes the structure of the soundtrack, recognizes the speakers and generates keywords as part of speech recognition. While we work with files that have already been created, the speech recognition of the “Fraunhofer Live Recognizer” even enables transcription in real time, so that – as already implemented in the Saxon state parliament – a parliamentary speech is subtitled live and thus becomes barrier-free.
Another very practical innovation is realized in audio mining via automatic speaker recognition. This makes it possible to specifically search for specific original sounds and answer questions such as “What did who say about a certain topic?” The reliability is very high even for sequences of ten seconds or more.
Multilingualism in other languages is seen as a current challenge: While the automatic recognition of German and English is already quite advanced, work is now being done on additional languages in collaboration with Deutsche Welle. In addition, the focus is increasingly on the transcription of audios, in which a translation text is spoken over sounds in the original language. And when it comes to deviations from standard German, such as dialects or “Denglish” in youth language or technical terms, you can imagine the difficulties that will have to be overcome in the further development of automatic text recognition.
__________________________________________________________________________________
Highlight detection and video compilation
__________________________________________________________________________________
Jakob Rosinski from IBM Services presented a system for generating metadata that analyzes videos from the sports sector (tennis, football) based on the IBM AI “Watson”. On this basis, the best scenes and images of a match, i.e. the highlights, are compiled fully automatically.
Based on machine learning, this system is able, among other things, to determine the beginning and end of a scene, statistically evaluate ball positions and speeds and discover sequences with a high level of excitement. The basis for this sentiment analysis are cheers in the audience, raised arms of the players, but also the changed language mode of commentators, which clearly characterizes the key moments of a competition. Another feature is the change in camera angles according to given rules: While the classic perspective predominates in the normal flow of the game, outstanding scenes are filmed in close-ups or from the goal perspective. By ranking this rather specific metadata, the system can independently put together two to three minute highlight sequences.
Short videos like these can then be made available for selection via a frontend. The personalized selection possible here (“Interesting or not?”) goes back to the system via a return channel for analysis and machine learning. This is where the foundations for truly individualized television, “TV 3.0”, emerge.
Jakob Rosinski named the topic of “re-broadcasting” as a current development: The AI automatically recognizes and removes the “brandings” at the beginning and end of videos from other broadcasters that already exist and are to be further used – and this almost in real time.
In response to a question from the expert audience about what the use of AI will mean for the future of media archivists, the speaker replied that in his opinion they will not become obsolete, but will have a changed, “orchestrating” role in the future, for example in the process of automating the media Adopt keywording.
__________________________________________________________________________________
VIVA – Visual information search in video archives
__________________________________________________________________________________
Joanna Bars from the German Broadcasting Archive (DRA), Nikolaus Korfhage (Philipps University of Marburg) and Kader Pustu-Iren from the TIB Hannover presented the research project VIVA for the conception and development of a video mining and retrieval system. The DRA’s inventory of GDR television program parts serves as the source material; the software developed is open source. Part of the system is a module in which training materials are annotated and enriched by documentarians.
For example, facial recognition is implemented using neural networks (Convolutional Neural Network, CNN) and thus machine learning and AI, which proves to be particularly suitable for semantic content. The average precision is given as 81%.
Machine training was briefly demonstrated. An existing individual concept, in the example given a “bakery”, with already labeled images was applied to additional images from a different corpus and subjected to a similarity comparison. This led to the further development of the individual concept and was in turn incorporated into the overall concept. The results of this project, including the similarity search, will be integrated into the common ARD television database “Fesad” and also allow intuitive research options there. This will make retro-digitized video material easier and more accessible in the future.
And the competition from commercial providers? The team responded to this question by saying they had researched something similar on the market and found nothing.
__________________________________________________________________________________
AI-generated metadata – From technology observation to productive use
__________________________________________________________________________________
Dr. Dirk Maroni, information manager at Westdeutscher Rundfunk, explained the AI activities at WDR for content analysis. These can be divided into technological and product-related aspects. The technological basis included data analytics as a decision-making tool, for example for format development, as well as content analytics for generating metadata from videos, audios and texts. On the product side, AI is used to implement accessibility and to synthetically create media (videos and audios). In addition, AI serves as the basis for robot journalism for the generation of texts and for the use of voice assistance systems.
When using artificial intelligence, WDR proceeds in five steps. Initially, the market is looking for AI products among start-ups and academia. In the second step, these are assessed for their specific usability. The next step involves validating the “experience” of a product, i.e. how it “feels” to potential users such as editors. In the fourth step, the overall process is evaluated, and the fifth step is finally the commissioning of the AI system.
Maroni then presented the WDR KIMD prototype for AI-generated metadata. He explained the “experience evaluation” step and demonstrated the text-to-speech transformation using the example of the landing page of the vfm conference. The system automatically recognizes people, speakers, organizations and locations in videos. Despite all these advances, one should not forget, said the speaker, that AI is a statistical process and is therefore inevitably subject to errors.
Another area of application is the “ARD Crossmedia Search” on the ARD mining platform for content research over a hundred million objects in the form of audios, videos, music and press releases. Using the example of the “Vendeé Globe” sailing regatta, the search and filtering of keywords and the search for videos that contain certain original sounds were demonstrated.
Dr. Maroni emphasizes the accompanying involvement of users (especially editorial teams) in the use of the technologies. This would have formulated a crucial point that was probably important to all speakers in this session without it being specifically stated: the AI technologies described are tools for a future in information science that have already proven their practical suitability today.
LexisNexis Risk Solutions
Global Spend on Financial Crime Compliance Reaches US$214 Billion
(BIIA) LexisNexis Risk Solutions has released its annual True Cost of Financial Crime Compliance Global Report. The results derived from the comprehensive survey of 1,015 financial crime compliance decision makers at financial institutions including banks as well as investment, asset management and insurance firms globally. The projected total cost of financial crime compliance across all financial institutions reached $213.9 billion in 2021, surpassing the $180.9 billion recorded in 2020.
The majority of this sizeable year-over-year increase is represented by Western Europe and the United States. The decision makers who took part in the study oversee financial crime compliance processes such as sanctions monitoring, know your customer (KYC) remediation, anti-money laundering (AML) and transaction monitoring. Key findings from the LexisNexis Risk SolutionsTrue Cost of Financial Crime Compliance Global Report:
- Western Countries Continue to Spend Highest on Compliance –Western European countries and the US continue to represent 82.7% of global total projected costs. Germany and the US bear the bulk of cost increases at $9.6 billion and $8.8 billion respectively with Germany outsizing all other countries by a considerable amount. Mid to large financial institutions lead this growth where all regions, excluding South Africa and the Middle East, show double digit percentage increases in compliance costs.
- Less Consensus on Operational Challenges –In previous years there has been consensus on the top two or three ranked compliance challenges within financial institutions. There is less uniformity in this year’s survey. Customer risk profiling, sanctions screening, regulatory reporting, identifying politically exposed persons (PEPs), KYC for account onboarding and efficient alerts resolution are all similarly ranked as key challenges. Different regions see varying degrees to which certain challenges are more heightened, however.
- Pandemic Impact –The ongoing pandemic has left a significant imprint on compliance departments, which exacerbated existing issues and led to an increase in the time and spending needed for due diligence. Mid and large firms in the US and Canada and parts of LATAM experienced sizeable pandemic-related cost increases. Key operational challenges have become increased in these markets since the start of the pandemic, including increased alert volumes and suspicious transactions, inefficiencies with alert resolution and due diligence, more manual work and limitations with proper risk profiling/sanctions screening/PEP identification.
- Technology Investment Leads to Better Outcomes –Financial institutions implementing technology solutions to support financial crime compliance efforts have been more prepared and less impacted overall by increasing regulatory pressures and COVID-19. Compared to firms that distributed more of their annual compliance costs to labor, those that allocated costs more toward technology are seeing smaller year-on-year financial crime compliance operations cost increases, lower costs per full-time employee and fewer pandemic-related challenges.
Business Information Industry Association (Hong Kong) is the international partner of Open Password
dpa
Fact checking: Against fake news and disinformation
(dpa) The German Press Agency shares its knowledge and has published a new white paper. The title “Factchecking – Journalism in the fight against lies and forgeries” deals with the topic of verification as one of the central editorial tasks. The focus is on know-how for debunking false information and tools for digital research. The eleven-page white paper can be downloaded from the dpa website .
A dpa-owned survey in which more than 80 editors-in-chief took part shows that two thirds of them now see false claims and manipulation as a serious threat. Over the past few years, the dpa has developed in-depth fact-checking expertise and also passes this on in training courses and other formats.
For example, as part of the Fact Check21 project , editors from more than a hundred media companies are currently deepening their knowledge of how false claims, disinformation and propaganda can be recognized and refuted with the help of the dpa. In addition, around three months before the federal election, the dpa developed a free training platform that explains techniques for digital research, verification and fact checking and provides many tips and useful links ( dpa-factify.com ). Factcheck21 is financially supported by the Google News Initiative.
Contents of the white paper: “Checking facts is the central task of journalism” – Alexandra Borchardt, journalist, consultant and professor for media change (Berlin University of the Arts, Hamburg Media School) – “Fact checking today and tomorrow” – Baybars Örsek, Director of International Fact- Checking Network IFCN – Know-how part: Basic knowledge verification – Training platform, training courses, fact-checking community – – New fact-checking formats.
OpenPassword
Forum and news
for the information industry
in German-speaking countries
New editions of Open Password appear four times a week.
If you would like to subscribe to the email service free of charge, please register at www.password-online.de.
The current edition of Open Password can be accessed immediately after it appears on the web. www.password-online.de/archiv. This also applies to all previously published editions.
International Cooperation Partner:
Outsell (London)
Business Industry Information Association/BIIA (Hong Kong)
Open Password Archive – Publications
OPEN PASSWORD ARCHIVE
DATA JOURNALISM
Handelsblatt’s Digital Reach