Artificial Intelligence & Machine Learning

Content is strategic for the success of any brand only if it is:

  • designed to be reused on multiple channels and devices, and for different purposes;
  • organized and archived in a coherent manner;
  • tagged systematically .

Often, however, this does not happen. Brands fail to manage the quantity of assets produced, which becomes a threat for the coherence of their messages and is responsible for poor customer experiences.

Marketing departments are called upon to confront the situation with resources that are often insufficient. They therefore settle for purchasing new content at the lowest possible price, at the expense of the quality that is offered to the user.

Applied to content, artificial intelligence is able to tame the chaos that reigns within companies and:

  • reconsolidate brand identity;
  • automate and speed up editorial processes;
  • avoid wasting resources, in terms of time and money.
Leaving the honor of organizing, applying tags and automatically analyzing each asset to artificial intelligence, marketers can finally concentrate on data analysis relating to the use of such content, and then on the definition and optimization of their own brand’s strategy.

FAQ & Info

How does the semantic analysis of documents in THRON work? 

The analysis of texts becomes ever more fine-tuned thanks to the constant evolution of computational linguistics.

Documents and texts go through a preliminary phase of stemming and tokenization, which is available in 11 languages.

The sections of the text that are relevant to the overall meaning of a specific document are set apart from those that are not.

The relevant textual elements are then ingested by various classifiers. At this point, the system also extracts those notions that cannot be derived directly from the words used to express them (resolving ambiguities in this way too), and that are not related to a specific language (indeed, the same concepts will be extracted from a document translated into different languages).

THRON, therefore 

  • recognizes the context and the concept expressed in a text
  • solves disambiguation
  • aggregates concepts expressed in different languages with the same notions
  • learns and assimilates new concepts that, from now on, it will be able to recognize.

Using this information, the semantic engine consolidates different notions that are related to the same conceptual sphere, identifying a topic. Subsequently, it automatically assigns relevant tags to each document, while taking into account your company’s dictionary tags.

Discover more

Can THRON semantically analyze images and videos too? 

The convolutional neural network lies at the base of THRON’s AI engines’ visual recognition technology.

Visual elements and patterns are extracted from images and are subsequently applied to models that identify objects and characteristics. These are the concepts that, when combined together, produce the tags that represent the image’s content.

With this same methodology, THRON recognizes and analyzes video content too (feature available by 2018).

Discover more

How can I personalize the way in which the semantic engine classifies digital assets? 

THRON’s users can define their own taxonomy in the Tag Center. Official taxonomy may be established by your own sales and marketing departments. This way, THRON will also tag in compliance with the company’s strategy.

If a relationship is set between tags, this will also be automatically transferred to the corresponding concepts.

The tags that make up the taxonomy can be structured:

  • into heritage relationships, or rather, where some characteristics, such as personalized metadata, are transferred from parent tags to the related children;
  • into combined relationships, or rather, where a tag is inserted into another, taking with it all the content that it is associated with.

Is it possible to search for content “by concept” or “by characteristic”? 

Of course. This too is one of the advantages offered by THRON’s AI engines: the ability to search content by subject matter. For example, you may carry out a generic search for “wedding images”, or documents that have “microeconomics” as their topic. An autocompletion mechanism in the concept dictionary allows users to easily tap into all THRON’s knowledge.

Does THRON understand the behavior of end users?

THRON recognizes the interests of each individual, analyzing the subject matters relating to the content that the company publishes online, and that is then viewed by each and every user.

This can be achieved because THRON applies the relevant tags to each digital asset. In this way, it is possible to discover what the user who has consulted specific content is interested in.

Thanks to its behavioral engine, THRON produces a synthetic and exhaustive representation of all the interests demonstrated by a particular contact. The tags are linked to those users who have consulted content with the same tags, and the frequency of interaction between the two is then measured.

Clearly, some concepts may be overrepresented, for example “car” if we are in a car company. To solve this problem, THRON’s algorithm also takes into account the tags’ rarity index.

Another aspect to bear in mind is that the importance of a specific subject matter may change for an individual over time. For this reason, the algorithm for behavioral analysis grants less weight to past events compared with recent ones.

Behaviors and taxonomies are constantly evolving. How does THRON manage to adapt? 

User behavior is dynamic by nature: it changes according to trends, habits, and due to external influences. THRON takes into account such dynamism in its creation of contact profiles or when it proposes content to them.

Taxonomies are also dynamic. THRON allows you to update your taxonomies at any moment. When a change is made to the taxonomy, THRON propagates it back through the entire history of each user, consequently updating profiles and adjusting suggestions.

In this way, the taxonomy approach is incremental. It builds itself up step by step, understanding as it works over time which data is most relevant for the company.

What criteria does THRON’s Predictive Content Recommendation use to suggest content?

THRON’s content recommendation engine adopts a highly evolved Machine Learning approach. It does not make suggestions that are based simply on similar tags, but combines the interests of the user, content topics and the interactions that users have with content and channels in real time.

This approach is considered “hybrid” because it combines content-based techniques and collaborative techniques. In this way, it reduces the problem that arises when there is only a small amount of data available on which to base recommendations.

In the content-based approach, content and contacts are represented by a series of characteristics (subject matters and interests) from which an indication of similarity is derived. In this instance, the recommendation is based on the idea that the user will like content that has characteristics corresponding with their interests.

In the collaborative approach, all actions performed by the user with regard to specific content are interpreted in the form of a “history”. Tracking of these past actions may then be analyzed to predict what the user might like, based on what others liked before them.

Discover more

Want to try THRON's Artificial Intelligence
applied to your company?
Contact our specialist!