Q&A WITH THE CTO

By Anne-Mette Riisgaard

I sat down with Hypefactors CTO, Viet Yen Nguyen, to discuss the technical benefits and challenges of the
Hypefactors platform.

On a cold spring afternoon in March of 2022, I sat down with Viet Yen Nguyen, CTO of Hypefactors, to obtain a deeper understanding of what an AI is, how it functions, as well as the ways in which the Hypefactors platform helps clients navigate through the vast jungle that is our contemporary media landscape.

This is what Viet had to say:

Tell me a little bit about yourself:

“My name is Viet and I am the CTO of Hypefactors. I have been educated as both a scientist and as an engineer, specializing in combining computing, probabilistic and temporal logic. The first decade of my career was in the space and automotive industry – I also obtained my PhD in these fields. Even though I was schooled by the theoretical branch of academia, my desire for making theories work in practice has always made me keep a firm leg on their applications. Over the years I moved deeper into the application domain. When I came across what kind of journey Casper (founder and CEO) had started with Hypefactors, I was attracted to join to help innovate and develop the technology to become second-to-none. That is why I moved with my family from Germany to Copenhagen.”

Why should people pay attention to Hypefactors? What are the company’s strengths compared to your competitors?

“We are the real deal when it comes to AI applied to media intelligence. We have made it our business to deploy task-specific AIs, each one able to perform one aspect of what a human intelligence would be able to. Our strength lies in the marriage of scientific inquiry and engineering methods. A marriage that brings us continuously at the forefront with our clients where theory leads to new applications, and new applications inspire new theories.

What’s the deal with AIs?

“Historically speaking, an AI has been defined as a building system whose interactions are indistinguishable from an actual human being. The thing is, though, that such a single AI does not exist. And there are no findings reported yet – not in academia nor in industry – that directly links today’s tech and theories to such a future entity. Today’s best tech is creating AIs each optimized for a very specific task.”

Why are AIs so important for media intelligence?

“Our contemporary media landscape and, by extension, the very impact that it has, is evolving and growing faster than ever before. This holds both for trusting and truthful narratives, as well as harmful and fake narratives. In order to keep up with the constant flow of information, we need to be able to decipher data from all over the world. The magnitude of data processing involved can only be effectively performed by AI.”

How does an AI actually work?

“The design of state of the art AIs are inspired by the human brain. Contemporary AIs are artificial neural networks made to resemble the neurons and their connections in a human brain. Imagine, if you will, the brain of a child. The neurons of a baby’s brain do not exhibit any refined cognitive skill. Over time, the child must be taught to know the difference between right and wrong; it must be taught specific inputs for specific desirable outputs and undesirable outcomes and vice versa. We do the same with artificial neural networks, albeit at an accelerated pace. Essentially, artificial neural networks are based on linear algebra. Inputs are transformed into a stream of numbers that themselves are represented as streams of bits (0 or 1, ed.) Utilizing linear algebraic techniques, the artificial neural network transforms inputs into an output, which by itself is a stream of numbers. These numbers are then mapped to interpretation and displayed to the user. This, for example, allows you to see if a mention is positive or negative.

What are the most important AIs? What do they do?

“In my view, there are two – AIs for big data collection and AIs for measuring media impact. While ‘big data collection’ is pretty self-explanatory, ‘measuring media impact’ is a bit more complex because it includes factors like trust, reputation, reach, prominence.”

What are some of the benefits of having an AI?

“There are a lot of benefits! First of all, an AI performs superiorly to a normal work force because it is not limited by human capabilities. For example, while a well-educated person might speak 4-5 languages proficiently, an AI can decipher 70+ languages. In other words, an AI can be used to reveal insights from media sources that otherwise would have been unavailable due to geographical or language barriers. Second of all, and this is very important, AIs are designed to perform systematically over time. This does not hold true for people. A person’s judgment may vary due to e.g. fatigue or external influences, whereas an AI will provide the same assessment of the same information every time. In this way, an AI can provide helpful and unbiased information that you can use in your favor as a baseline – for example in a quarter-to-quarter comparison or if you need to compare two different brands against each other.”

Can’t you just use Google instead of the Hypefactors platform?

“No. Not if you want to yield the best results. Google only reveals snippets of information that do not give you an overview of the media landscape. You do not know the relevance of Google’s search results either. In contrast to Google, the Hypefactors platform can predict when a new article is published on a particular blog, and then crawl it immediately when it becomes available. We also show how impactful the harvested information is in terms of reputation, reach and relevance.”

What are the specific AIs that comprise the Hypefactors platform? How would you describe them?

While the devil is in the details, we typically group the AIs by their application. This brings about six categories of AIs that work together. We refer to them as »Source-AI«, »Understand-AI«, »Perception-AI«, »Origin-AI«, »Context-AI« and »Character-AI«.

• Firstly, our »Source-AI« comprises algorithms that discover and maximize data intake. They vary from section detection algorithms and optimizers of the crawling schedule. They ensure that you get the best and most comprehensive external view of your media monitoring.
• Secondly, our »Understand-AI« transform the raw unprocessed data from the »Source-AI« into a structured and machine readable format. It includes algorithms that identify and interpret the publication date, the headline and article text, but also OCR and transcribing spoken dialogue from TV and radio broadcasts.
• Thirdly, »Perception-AI« analyzes the media content on how readers would perceive it. It uses deep-learning based natural language processing to identify the prominence and reputation.
• Fourthly, our »Origin-AI« analyzes where the media content is from. It includes language and country identification analysis, and also identifies the newspaper, blog or influencer and whether they are a specialist outlet or catering to a broad audience.
• Fiftly, our »Context-AI« digests the data coming from »Understand-AI« and digests it into summaries, extracts main keywords and determines the most fitting categories. Furthermore, in Q2 2022 we will launch NER, which will identify products, companies, locations and persons from the text.
• Sixthly, and lastly, our »Character-AI« dives into the media type and its format. The algorithms here identify length, duration (if applicable) and its presentation: like whether the contents is self-promoting or written in the style by an independent journalist. Or whether it’s a caption or a byline. There is more here that we have on the roadmap to crack, like identifying irony, satire and generated content.

I noticed you used the word “special” to describe your media crawler. What is it about the Hypefactors media crawler that makes it so special?

“There are two elements to it. The first is scale. We are one of the few companies in the world that can crawl millions of websites while also maintaining data freshness. This requires in-depth large-scale systems engineering for which we have all engineering know-how in house. The second element is that it is of relatively modern design, meaning that the crawler is designed to utilize AI to crawl faster and more precisely. For example – we have an AI that crawls text as well as navigation links for websites, making it possible to crawl deeper and more effectively. Moreover, we have a publication-date AI which can determine the publication date of any article regardless of language.”

How much data do you process?

“We do over 2 billion crawler movements each month, which yields into 200 million
mentions found each month.”

What do you think the future of media intelligence and reputation management looks like?

“The media represents the collective thoughts that people find desirable of receiving attention. Our periphery is getting wider – we are, in a sense, opening the world. Up until recently, languages and distances were natural barriers that hindered us. But technology is now breaking those barriers, and we find ourselves riding that wave, bridging the gap between languages and cultures. And we are only at the very beginning. I expect that we are only going to unlock more insights as technology continues to improve.

And, finally, what do you think will remain unchanged in the next 10 years?

“Our client’s needs and the reasons behind them today will carry through the next decade and the decades after.

Clients need to track their product launches, track their baseline reputation and compare that to their competitors, identify problematic suppliers and find alternatives to them, and overall gauge how key strategic decisions are received. We also know users want to have an easy to use platform and they want to have these insights fast.

And we are only talking about today’s needs and wants. The range of clients’ needs to cater to will only grow and last. This is because advances in data and AI technology will enable solutions for needs that people are today unaware of, or ignore because of today’s technological infeasibility.”