From mass media to smart media: Noowit is using Artificial Intelligence to personalize your web experience

Por ~ Publicado el 29 junio 2016

What happens when you combine the theory of autopoiesis, research on immune systems and custom web design? A Greek researcher and entrepreneur may have a clue. Nikolaos Nanas spoke with Puroperiodismo about Noowit, a publishing platform powered by Artificial Intelligence. For him personalization, design and monetization are the future of publishing. This is his account of what Noowit has achieved so far.

Nikolaos Nanas

Nikolaos Nanas


» My PhD research focused on Biologically Inspired AI and in particular the application of Artificial Immune Systems for Adaptive Information Filtering. The latter is the field concerned with the problem of information overload on the web and is trying to tackle it by building interest profiles for each individual user, capable of distilling relevant from non relevant information. This is a complex computational task because people have multiple and evolving interests. In my research I argued that approaches like Neural Networks, Genetic Algorithms and in general, any approach that is essentially based on vector representations of a user’s interests is not well suited to the dynamic and multidimensional nature of the problem.


» I proposed an alternative computational solution, inspired by how the inner workings of the biologically immune systems are explained by the theory of Autopoiesis. The immune system has the ability to discriminate between the host organism’s own molecules and any external pathogen or virus. This is called self—non-self discrimination and is essential for the organism’s survival. When the discrimination fails the immune system attacks the organism’s own molecules causing autoimmune diseases. The theory of Autopoiesis has been developed by the Chilean biologists Humberto Maturana and Francisco Varela in their effort to provide a common characterization of the organization of living systems. According to the theory, the immune system is a network of interacting antibodies that constantly self-organises to define and preserve the organism’s identity, in what is called “self assertion”. As long as the immune system can maintain “self”, even while the organism developing within an ever changing environment, anything else can be discriminated as “non-self”.


» I developed a model, which I called “Nootropia” (Greek word for: “An Individual’s or a group’s particular way of thinking, someone’s characteristics of intellect and perception”). Instead of a network of antibodies, it uses a self-organising network of weighted features (keywords, tags, etc.) to represent and constantly adapt to someone’s multiple and changing interests, his ever developing “info-self”. The network is neither predefined nor fixed. It can both learn and forget. This immune inspired approach has two significant advantages: a) it can capture someone’s multiple interests by including thousand of features without suffering the curse of dimensionality, which is characteristic of the vector space mode and b) it can continuously learn without getting stack in local optima and without becoming saturated. In practice, this means that we can capture an individual’s or a group’s complex and dynamic interests with a single, autonomous computational structure, that can even run on the client’s machine for efficiency and scalability. Comparative experiments have demonstrated that the network structure, with its links measuring correlations between features, can be 50% more accurate that the popular vector-space representation. It avoids synonymity and polysemy issues and can identify silent features that capture aspects of what makes something interesting that even he could not possibly express. It “realized” for example that I like “murals” and more generally “street art” although I did not even know the word “murals” in English and if someone was to ask me what type of art you like I would never think of answering “street art”. The same happened when it discovered that I like surrealism in photography. Identifying my favorite sport team is basic.



» The crisis has seriously influenced the Greek Universities and Research Institutes. Job openings and funding have become obsolete. It made taking the risk of entrepreneurship easier. Personally, I got to the point where I felt that I could no longer contribute academically. I have published several papers demonstrating the computational and practical advantages of Nootropia, but they ended up just adding another line in my CV. I have also failed to attract research funding from the EU. I wanted to apply Nootropia to the world. We have developed and tested various research prototypes with my team at the Institute of Research and Technology of Thessaly (IRETETH) and in 2009 we had already developed possibly the first Personalized News Aggregator in the world, but it was just a guinea pig, miles away from a usable and valuable product. In 2010 I have visited New York and had some intriguing discussions with prominent web entrepreneurs, who reacted very positively and encouraging to my demos. This trip prompted me to turn Nootropia into a real world application. Back then, I did not even know what a startup was. I was still an academic. I just wanted to contribute to the web and I did not care about making money.


» It took me a year to find the right partners. We started developing Noowit’s Personalized News Aggregator in September 2011 with Christos Spiliopoulos and Thanos Theodoridis (who later left the team). The first private beta version was ready in May 2012 and the first public beta version was released in 1st of July 2013. Several versions and additional features followed (including the ability to publish your own online magazine) and Noowit attracted international interest, being considered by many a competitor to the well established Flipboard. By May 2015 however, having failed to attract a critical mass of users and funding and after realizing that the domain of News Aggregators is generally failing we have decided to change research direction. In June 2015 we have won the “Best Personalization Service” prize at the Digital Media Awards and this gave us the opportunity to directly speak with publishers and identify their needs. We decided to stop the further development of our Personalized News Aggregator and turn all our energy towards the development of an intelligent publishing platform. The new platform and our official pivot was announced last Easter. Noowit has become the “printing house” of the future for publishing intelligent/personalized newspapers and magazines online. Today we have already signed our first contracts and we our proud for our first four publications. The latest is OZONWeb and if you ask me, this is the first online magazine ever, the way it should be.


» Nootropia, lies at the core of everything Noowit does. Every article is evaluated according to the interests of each individual. The result is a relevance score which allows us to choose the most interesting articles and also, to appropriately present them to the reader. We have developed an intelligent layout engine to produce newspapers and magazines that adapt to the reader’s interests, the content’s volume and characteristics and of course the device being used. Each section’s layout is hierarchical. Articles relevant to the reader appear first and large and the least relevant articles follow smaller and smaller. So even if and edition’s content is common for all readers (e.g., if it explicitly selected by the editor and not distilled by our algorithms), the way it appears to each individual is different, to reflect his interests and help him identify the most relevant content. Our statistics show that this way of making the relevant content stand out does work. Readers are more likely to click, share and provide feedback to the recommended content.

» Web users become more and more saturated by the available information glut. Web giant has found a way to make it easier to find what you are looking for (Google), or to find popular-recommended content (Facebook). It is an essential part of their success. Personalization becomes more and more essential and an integral part of the web experience, even when it is subtle. It does not only make it easier and fast to browse through content, it creates engagement, which is essential in today’s competitive web. When I read The Daily or The Tag news about my team will appear first and large and those about our competitors last and small. It makes me fell good, as if the publication itself is supporting my team. Of course the opposite happens to the reader’s supporting the opponents. I know what is happening in the background, but they do not. They are just left with the nice feeling this is a newspaper or magazine that knows them and respects them.



» All known personalization engines out there are heavily based on constructing ontologies or concept hierarchies to capture the essence of a topic. The process of building and applying this type of structure heavily involves Natural Language Specific techniques which are language specific. It is not accidental all well established personalization engines like Zite (acquired by Flipboard), and others only work for English. In contrast, Nootropia is fully statistical. The only component required to apply it to a language is its stemmer for reducing words to their root (and even that could be avoided). Such stemmers are readily available off-the-shelf for almost every language. We already support ten of the most popular languages (including Spanish of course) and we can easily ad any other language. In fact, Nootropia is not only language agnostic, it is also domain independent. It could be applied to any type of information, as long as there are features describing its content or usage.

» One of the main arguments of my research has been that ontology driven approaches are not well suited to web personalization, due to their centralized, monolithic and non-dynamic nature. A common ontology or concept hierarchy can not express the different ways people conceptualize the world. It is characteristic that these approaches suffer by issues of polysemy. Try adding the topic “python” to Flipboard/Zite. It will aggregate articles about both python the snake and python the programming language. Nootropia builds a separate concept network for each individual reader and captures the specific term correlations that make something interesting to this particular reader. It can learn that it is python the programming language that he is interested in and even recognize the specifics aspects of python programming he is currently into (e.g., python programming for machine learning).



» One of our main goals with Noowit’s new publishing platform is to combine Artificial Intelligence, with innovative design and cutting edge technology to successfully transfer the effectiveness of print media online. When you open a newspaper or magazine you flip through its pages from beginning to end. You have selected this particular newspaper/magazine because you trust it, you enjoy its content and you like it for its typography, imagery and layout. You find interesting articles, but also serendipitous. Articles which are not necessarily directly related to your interests, but are still appropriately chosen and edited to compose a package of information. As you flip through this package you are exposed to full page ads without being obstructed and without having to change your flipping behavior.

» Online, newspapers and magazines suffered something similar to the music album. They broke into individual articles, which are disseminated and consumed individually via social media. Portals are by no means a counterpart to print media. They have no beginning or end. There goal was to lure the visitor in via SEO and then keep him in, within a maze of categories, subcategories and tags. Their Key Performance Indicator (KPI) is pageviews and average time on site. But this not what is happening any longer. Seventy percent of their visitors come from social media. They click on an article’s post, visit its page, read it and leave.

» One of our greatest lessons from our first Personalized News Aggregator was that the web will never move on to the era of the “page”. It will remain in the era of the “papyrus”, cause scrolling is simply the most fluent way to consume content online. The success of Facebook and Twitter, where millions consume terabytes of information, is a solid, indisputable proof. Noowit’s Personalized News Aggregator was possibly the first web app with paginated content and it took a lot of ingenious engineering to be achieved. It proved wrong and not user friendly. The new publishing platform takes a completely different stance. Newspapers and magazines become a papyrus with a beginning and an end that the reader “flips” through them by naturally scrolling on any device. They have a cover and a back page and in between a series of sections. Various algorithms are responsible, not only for the aforementioned personalized layouts, but also for aligning images and text to produce a result that is clearly reminiscent of a traditional newspaper/magazine only without pages. Visitors coming from social media are exposed to the full issue and not individual articles. The goal is no longer to keep the visitor in long enough to produce pageviews. The KPI is completely different. As long as a reader has browsed the full newspaper/magazine from beginning to the end its goal has been achieved. It can be checked in the bin. In the process the reader is exposed to a number of full page ads without being obstructed, like in a traditional print newspaper or magazine. Our readers find this new type of online ads pleasant and their Click Through Rates (CTRs) are over 4% when the corresponding CTR of the average banner is 0,06%. Noowit’s goal is to publish online newspapers and magazines the way they should be. With both the smarts and the looks.



» When people hear about Artificial Intelligence, the first thing that comes in their mind is humanoid robots with natural language skills and pseudo-feelings or at least a character of some sort. Artificial Intelligence is already here. From small bots to larger more complex AI infrastructure they are involved in the way content is produced, disseminated, analyzed, presented… Each of these AI entities individually have nothing to do with the intelligent robots people imagine. Individually, they can be as dumb as an ant, but collectively play a catalytic role in how we interact with information on the web. They are an integral part of web’s evolution and of its global effect on our lives. They are and become more and more part of the equation that defines human life on earth.

» If I focus on media in particular I will have once more to use an excerpt from Steven Johnson’s book EMERGENCE: “Amazingly, this process has come full circle. Hundreds of thousands—if not millions— of years ago, our brains developed a feedback mechanism that enabled them to construct theories of other minds. Today, we are beginning to create software applications that are capable of developing a theory of our minds. All those fluid, self-organizing programs tracking our tastes and interests, and measuring them against the behavior of larger populations, these programs are the beginning of a progression that will, in a matter of years, lead to a world where we regularly interact with media that seems to know us in some fundamental way. Software will recognize our habits, anticipate our needs, adapt to our changing moods. The first generation of emergent software displayed a captivatingly organic quality; they seemed more like life-forms than the sterile instruction sets and command lines of early code. The next generation will take that organic feel one step further: the new software will use the tools of self-organization to build models of our own mental states. These programs won’t be self-aware, and they won’t pass any Turing tests, but they will make the media experience we’ve grown accustomed to seem autistic in comparison. They will be mind readers.”

» With Noowit we want to contribute towards this direction. To turn mass media into smart, personalized media. Media Made Smart!