What if your asset management system could read between the lines?

Jedrzej Osinski, Ph.D.
4 min readJan 3, 2022


Photo by Héctor J. Rivas on Unsplash

A boosted and more intuitive asset search tool means more efficient and cost-effective digital content management. Check the teaser for our unique solution and the all-new user experience we have been working on at Wunderman Thompson Technology.

Believe it or not, but the language you speak is your very specific way of communicating; it’s just as unique as your fingerprints. In fact, your language is a one-of-a-kind result and reflection of your entire life: It depends on your social and ethnic background, your education, the society you live in, work specialization, mental and physical condition, and overall experience collected since your childhood. We look at things from different perspectives and often name the same things differently.

The complexity of natural language surely enriches our lives, but at the same time becomes a significant challenge for modern digital asset management. If you imagine a big content entry team, which, especially nowadays, is also highly distributed across various locations and cultural backgrounds, language ambiguity may generate significant costs and efforts.

Let’s start with a simple example and assume we are looking for an illustration of a car within our massive asset collection. Using “car” as a search term makes perfect sense, but it would not be that helpful if the images are named “auto” or “motorcar.” Maybe it was tagged as a “limousine” or just a “vehicle”? You can try and fail many times until you find what you want. So, is it possible for a system to read between these lines and make the effort instead of you? The answer is “Yes.”

To achieve this, we make use of the Princeton WordNet®, the world biggest lexical database of English, that has been in continuous development for over thirty years. Not only does it provide synonyms for almost every existing noun, but it also links their meanings with other words in a tree-like structure. This allows us to automatically identify what concepts are more general or detailed compared with others. Therefore, the system immediately knows that a lion is a mammal and a kind of animal, a car is a type of wheeled vehicle like a train, or even… that the Siege Perilous is a type of seat at King Arthur’s Round Table. Simply speaking, WordNet not only helps to understand the words, it also explains the world around us.

Let’s have a look at our WordNet-based asset search solution, whose initial code name was Okapi (as no-one is sure whether it is more giraffe or zebra). What’s worth highlighting is that the mechanism works independently on asset metadata and filenames without needing any AI-based image recognition. Suppose you look for a “car” — without our extension we just get all assets named “car”:

However, you can also ask Okapi for synonyms:

When you look for a “dog”, you get few assets from our collection; however, you can use our generalization mechanism (a broader term search) to look for other carnivores, mammals, or animals: Users can define ad-hoc how deep they want to dive into the semantic tree:

One can also extend the search in the other direction and look for “dog” but also enable the narrower term search to include specific breeds of dogs in the results:

Using these features to extend searches, one can easily find either more or more appropriate assets with little or no effort. Finally, the solution provides an entirely new and unique visualization of the asset collection. In addition to the standard list or card views found in many DAM systems, users can browse the asset collection via a tree view of semantically-related terms. This could be an invaluable tool for asset managers, for example, by highlighting areas of strength and weakness in the collection or assisting in the design of corporate taxonomies:

Tree-like semantic navigation across your digital assets

Our solution can significantly enhance existing asset search mechanisms. It can help reduce the time needed to find assets and mitigate cultural and linguistic differences when a distributed team manages huge collections of assets. The tool can also enhance diversity and inclusion by supporting users for whom English is not a native language. Finally, image search is really just the tip of the iceberg. If you consider that a document is also a type of asset, then we immediately see further applications: For example, this Wordnet-based mechanism could be used for various automated health-checks, including the structure verification of documents.

For more tech details, check the WTT Tech Blog.



Jedrzej Osinski, Ph.D.

QA Engineer, AI specialist, writer and speaker. Author of “Cunning Machines”.