If the writing style of the posts at ArcticStartup.com would be processed and analyzed by an AI algorithm, what would be the outcome?  Read on and you will find out the answer and get an update about interesting Swedish startups that are using AI techniques to improve the automated understanding of the ever-growing web content out there. This is the future so be sure to get a grasp of the basics right now.

This kind of automatically processed information for your website is all possible and available for free (at the moment and presumably also in the future) thanks to the Stockholm based Uclassify.com and its three team members, Jon Kågström (AI programmer), Emil Kågström (database/web expert) and Roger Karlsson (server expert). Jon Kågström quit his fulltime job as an AI programmer at EA Digital Illusions to work on the uClassify project.

That’s an impressive sign of dedication and commitment, isn’t it? That’s what startups are built of, complete dedication and confidence in the product and idea. The path to success,  no doubt. The uClassify team are looking for sponsors to enable them to develop the service further.

Typealyzer.com, one of the interesting services based on the uClassify AI engine, describes ArcticStartup’s  processed content and its providers in the following fashion:

The organizing and efficient type. They are especially attuned to setting goals and managing available resources to get the job done. Once they have made up their mind about something, it can be quite difficult to convince them otherwise. They listen to hard facts and can have a hard time accepting new or innovative ways of doing things. The Guardians are often happy working in highly structured work environments where everyone knows the rules of the job. They respect authority and are loyal team players.”

Next, we are provided with a graph displaying the active parts (assumed by the AI) of the ArcticStartup team’s brains while writing the processed content, as seen here:

Now contrast that to the AI claimed “facts” that can be extracted when running Techcrunch.com as the source of content.

“techcrunch.com is probably written by a male somewhere between 66-100 years old. The writing style is academic and happy most of the time.“

It continues describing the ArcticStartup equivalent for the US market in this way:

“The direct and assertive type. They are especially attuned to the big picture and how to get things done. They are talented strategic planners, but might come off as insensitive to others’ needs and appear arrogant. They like to be where the action is and like making bold and sweeping changes in complex situations. The Executives are happy when their work lets them learn and improve themselves and how things work around them. Not being very shy about expressing their ideas and often very outgoing they often make excellent public speakers.”

Many fields of application to the Bayes theorem (the AI part) are showcased by uClassify. Dimensions like age estimate, language detection, the tonality, mood, perceiving, judging, attitude, lifestyle and more can be accessed through the provided understandable open API .

Continuing our little comparison, we also discover that the ArcticStartup team has the mixed writing style of HG Wells/ J. Verne while Techcrunch has the style of HG Wells/O. Wilde/L. Caroll in that order of dominance. The system has been trained against three books by each author, collected from the Gutenberg Project.

How does a text classifier work?

To answer what category a text or website content is most likely to belong to, the AI learning function (a naive Bayes classifier) needs some training. Training involves creating all wanted classes (categories) and training each class on text documents that are of that type. In spite of the naive design and over-simplified assumptions, naive Bayes classifiers are known to work quite well in many complex real-world situations. Here is a Bayes theorem demo to shed some more light on this “magical” yet useful and simple formula.

How well does this AI approach work?

Well, it is hard to say based on some random searches, but the more feedback & training; the more likely it is the algorithm will hit the bullseye. Try for yourself and be your own judge of how efficient this method is and might become with some more tuning (read training). I think it works fairly well so far, but of course it is not perfect yet. Some of the functions are hard for us humans to really judge, which is why we would want to use these functions in the first place.

Other uClassify-based services to try out are urlai.com, ageanalyzer.com and bloggparti.se (Swedish).

To be fair, let’s present Aitellu.com and Saplo.com as well, to get a picture of the prominent startups in the text analysis segment in Sweden at the moment (If your company should have been mentioned, please drop me a line and I will do my best to find a remedy).

Aitellu.com – Since 2002 (not a startup anymore) they have used self learning systems to understand digital texts, in order to give their customers the option to act on that extracted valuable information.  All together Aitellu employs 11 people in Gothenburg with clients in the Nordic countries as well as in the U.K..  They are looking for additional sales staff at the moment (according to a recent Twitter message).

Saplo.com – They utilize semantic technologies to extract opinions and sentiment in text. They claim that their technique is equivalent to having thousands of employees constantly searching and sifting what is written about your organization. Saplo employs 7 people.

Related links:

No more articles