The future of Media intelligence
What is the future of media intelligence? Can artificial intelligence (AI) replace human touch in media reporting? Which tendencies and technologies will grab our attention in the next few years if we do not want to hop off the trend board? Can one keep up with the pace of permanent content expansion beyond the capacity of the human brain? These are only a handful of questions related to the constant evolution and creation process in the media intelligence industry.
“As a tech solution provider, our mission is to give answers to our customers,” Identrics CEO Vladimir Petkov emphasised at the second session of the FIBEP “Exploring the media intelligence business ecosystem” interview series. His personal belief is that in the long-term, Identrics, as well as any other tech solution provider in the media intelligence ecosystem, should automate as much as possible, otherwise it would not be future-proof and able to address the needs and challenges of today’s media environment. Social media, for example, produces huge amounts of documents, as much as half a million per day, which would be impossible to process solely with people or, at the very least, would not be feasible. “That is why technology is what makes the industry future-proof,” Vladimir Petkov commented.
Identrics is, therefore, focused on the in-house development of various AI technologies to automate as many processes as possible. With its expertise in automation extraction, Identrics collects web data, then automates knowledge extraction from millions of documents using all the languages and focusing on the major ones.
Connecting technologies such as machine learning, deep learning and AI to a real world problem could sometimes be a challenge. For Vladimir Petkov, technology is just a tool, but it is designed to solve real world problems. He believes there are three general areas of technology to keep an eye on in the next few years.
Automated abstractive summarisation
One of the extremely hot topics nowadays is text generation. Around 70% of Identrics’ research is now focused on automatic translation. There is, in fact, a working solution, which is to be marketed in approximately three months. “What we managed to do is to recreate the original document in a much shorter format,” Vladimir Petkov explained. The usual extractive summaries are produced by a simpler algorithm, which reuses and just glues together the most interesting sentences from a certain document. The optimisation achieved by Identrics, however, is aimed to analyse the same document, its topic, tone of voice and sentiment, and only then create a completely different and perfectly new text based on all these aspects.”
“What is changing in the industry is how we reach and how we deliver to our customers, and this technology allows delivering in real-time,” Vladimir Petkov stressed. As a product, summaries bring value to customers but they are just the beginning. With this text generation technology, there are a lot of opportunities to further automate report writing. By providing unique content, automated summaries also carry the potential to solve some of the copyright issues in regard to media reporting.
Another promising technology is, in his opinion, knowledge bases, or knowledge graphs, as they are called today. Knowledge bases use a graph-structured data model or topology to integrate data. They are, in fact, a very old technology that is currently gaining momentum.
In comparison to the typical databases, knowledge bases offer two main advantages. All knowledge that is stored via this technology can be reused. Non-obvious and hidden knowledge is also easily discoverable. Knowledge bases could then be integrated into semantic networks, which are networks between different knowledge bases.
If that happens, they can then be connected to public knowledge bases such as Google, which contains some five hundred billion facts for five billion entities, or Wikipedia. If so, one of the hardest problems in the machine learning and deep learning world could be solved. There comes the third hot technology, known as entity resolution.
Entity resolution and entity linking models allow perfect entity disambiguation, followed by linking each entity to the appropriate entity. Basically, this would mean recognising a certain person, organisation or structure among all other entities with an identical or similar name and related data and putting them into the right context.
Combining these three technologies would allow Identrics to create extremely deep reports and include non-obvious knowledge. This is something that should be put in production and pipelines to be integrated into the company’s system.
FIBEP & Identrics
Vladimir Petkov believes that Identrics’ FIBEP membership provides a unique chance for the company to keep up with the latest trends and developments in the media intelligence business and discover new cases to showcase their possessed technologies.
Most companies in the industry have similar entry problems, and workshops like FIBEP’sExploring the Media Intelligence Business Ecosystem interview series help these companies understand media intelligence better. Meetings, where competitors and partners present their results of proof of concept or hypotheses and report on their research and development pipelines, are needed on a more regular basis, Vladimir Petkov believes.
Business associations make a positive change in the field of expansion of standardisation efforts, such as XML, but there are things that could be improved too, such as training data creation and training data exchange. Last but not least, the constant updates of the regulatory framework in different regions worldwide should also be introduced, discussed and evaluated on a regular basis.