Artificial Intelligence Engines

Business Case:

KPI to be measured

Bookmark this resource Follow

Ask a question

Was this article helpful? 3 out of 3 found this helpful


THRON's Artificial Intelligence is a great help to editors in the content creation process.

By automating the tag application and content classification, it saves time and effort for editors so they can concentrate on the creative process and analysis of results.

THRON's artificial intelligence relies on two powerful engines:

The first is the Semantic Engine, which is activated as soon as each content is loaded.

The second, the Behavior Engine, continues its work over time.


The Semantic Engine understands and classifies all your content (texts, images, videos) and applies tags, both by creating new tags and on the basis of the official brand taxonomy.


The Behavior Engine constantly monitors users' access to your content and improves its classification based on the use that is actually made of it.

It allows you to refine and improve the profile of users, based on the topics they have consulted; and improve the recognition of the “Personas” that are interested in your content, drawing them from the profiles of the users who have been most engaged.


Semantic Engine

THRON comes out-of-the-box with a powerful semantic engine, capable of analyzing content the moment it is published into the Platform, automatically creating and assigning tags related to its topic. The elasticity of the engine in tag assignation is entirely customizable.

All tags extracted from the Semantic engine will be automatically translated into all available and active languages in the Dashboard. Automatic translation of keywords included in the iptc or xmp metadata of images and those included via meta tags inside the pages imported by Magic Site Integration is excluded.


The following languages are currently supported: English, Chinese, Dutch, French, German, Italian, Polish, Portugese, Russian, Spanish, Swedish.


Tags created by the Semantic engine are "not categorized", but can be manually introduced into the dictionary by Content Intelligence Managers. Semantic engine will not generate multiple tags with the same name. If a tag already exists, it will be automatically associated with the content. Replacing content's version will trigger the update of tags associated by the engine.


All the actions performed by the engine over tags (creation and association with content, removal from content) will only affect engine-generated tags unless they have been manually added to content.

Semantic engine is available for TOPIC classification only. It cannot be enabled on TARGET classification. It operates on the following content types:

  • Documents: tags will be assigned according to engine configuration and the number of times each word is repeated. Supported file types are: txt, html, doc, docx, pdf, ppt, pptx. Content title and description will be taken into account too. The semantic analysis of documents is more efficient if the actual text is long: this because, if the text is long enough, the artificial intelligence will  understand the actual scope of the content, thus contextualizing all the concepts therein present. If document includes text with multiple languages  concept extraction will be less accurate.
  • Images: tags will be assigned according to elements represented in the picture. Content title and description will be taken into account too. If the source file has some keywords in the form of IPTC metadata, those will become semantic tags.


This is a document about THRON CDN performances, processed by the engine with different configurations.




Have a look at the result:

These tags have been extracted with engine recognition reliability at 50%:

These tags have been extracted with engine recognition reliability at 60%:

These tags have been extracted with engine recognition reliability at 70%:

These tags have been extracted with engine recognition reliability at 80% and more:


Behavior Engine


THRON Artificial Intelligence leverages a set of algorithms for processing data on how content is used. It can correlate visitors, content, access frequency and other relevant data.


The main goal is to transform collected data into visualizations, thus generating insights, trends and other important information.


THRON users and content can be categorized and enriched by tags.


Thanks to its Artificial Intelligence, the THRON platform is constantly monitoring access to content, thus automatically improving its classification according to its actual use: comparing content classification with users’ interests. At the same time it will provide a better user classification by comparing their interests with the topics of the most visited content.


This mechanism will make you always capable of proposing content that is relevant for specific users.

Each asset is automatically enriched by THRON semantic analysis and image recognition with tags that describe the concepts and object present in the asset.

The THRON Behavior engine processes several user access parameters, including content view duration, content view frequency and recency. The Behavior engine works at concept level so it extracts, among the other values, how much time a user spent on a specific topic, how often he engages with that topic and how recent his interest towards a topic was. Based on those values, it enriches the user profile (works also on anonymous users) by adding topics of interest as tags.


The engine will remove tags managed by the engine ONLY. Tags that have been added manually will never be removed by the engine


Tags which have been automatically added by the engine cannot be removed manually (because they are based on real analytics).

The Behavior engine analyzes and collects data daily, but applies tags based on data collected once a week.




Below is an infographic that will help you understand the functioning of the Behavior Engine.


Let's suppose you have two profiled users, a woman and a man, and two contents, an image and a video. These four entities have already been tagged with the following set:


As soon as the two contents are embedded in your websites, your users start watching them. In particular, let's assume that the woman looks at the image content since it matches her interest, while the man looks at the video content for the same reason.



Over time, the Behavior engine will process this information and if the visits by these users to content exceed the configuration threshold, it will assign to content the tags of the TARGET class, comparing them with those of the users accessing such content.



Similarly, a feedback process will be carried out: if the visits by users to content exceed the configuration threshold, the Behavior engine will add tags in the users' TOPIC class, comparing them with those of the content they viewed.

In the Tag Center, under the configuration panel of the behavior engine, you can change its sensitivity to changes in users’ behavior. You can select how quick behavior should act, for example, from daily to monthly.

Was this article helpful?
3 out of 3 found this helpful

Have any question?

Open a ticket