The main function of DAM systems like OpenAsset is to simplify the process of applying metadata to files, and to enable users to easily access the files they need. A well-organised DAM ensures people spend less time retrieving images and more time creating impressive bid proposals and marketing collateral.
The development of the Image Similarity search feature, explored in our previous blog, made us aware of the opportunities offered by incorporating Artificial Intelligence (AI) and image recognition (IR) technology into OpenAsset. This feature offers an enhanced image search without the need for manual tagging, allowing users to discover assets with no extra administrative effort.
While this is a valuable service for users, our years of experience in the Architecture, Engineering and Construction (AEC), and Real Estate sectors tells us that keyword-based search will remain important to our customers. We know that certain searches will always start with commonly used industry terms, and it is by predicting these terms accurately and consistently that our software could deliver the most value. We have previously explained that using third-party image recognition services for auto-tagging is problematic due to the level of inaccuracy and the lack of industry specificity. We are keeping a keen eye on these services to understand if there are likely to be developments in the technology that would directly benefit the AEC and Real Estate industries in the near future.
OpenAsset has a configurable keyword taxonomy with a hierarchical structure that is used to standardise terminology across the DAM. We have found that restricting the keywords around an agreed set of terms is helpful for search as our customers can set the terminology that they commonly use. Although this approach is helpful in ensuring consistency in how assets are tagged, it’s clear that our customers would benefit from spending less time managing keywords. To create a solution for this in the near-term we have been exploring other ways to streamline the process of tagging assets using AI-generated data.
AEC and Real Estate Specific Keywords
In order to recommend relevant terms for our customers we must be clear on the keywords that are commonly used within these industries. We are well placed to understand this and by drawing on our industry experience, we have been able to analyze and understand our customer data with a high level of insight. From a corpus of around 15,000 keywords we have been able to identify a set of commonly used terms, which have formed the basis of our AEC and Real Estate specific keyword ontology. You can think of an ontology as a vocabulary of terms that are related or restricted by meaning.
Project and File Keywords
A significant benefit for OpenAsset over other DAMs is the project-oriented structure and the ability to apply keywords to projects as well as to individual files. The value of project keywords is that they are easier to manage, yet assist with both project- and file-level filtering. Our research has shown that customers spend significant amounts time managing project keywords, so this seemed like the most effective place to begin assisted tagging.
Project Keyword Suggestions
We settled on the idea of building a project keyword suggestion engine. This would use the data generated by the AI-powered image recognition service (Amazon Rekognition) via our keyword ontology in order to infer specific terms that we believe our customers are likely to find useful. An example of this feature in action would be if the image recognition service had analysed an image and recommended the term ‘Skyscraper’. We would suggest using the term ‘Highrise’ for a project keyword as this is a more widely used industry term. To test this engine we have built a lightweight app so we can assess recommendations across many scenarios. We are now at the stage where we can build this out as an experimental feature that we can improve incrementally based on user feedback and results.
Building the Feature
Our first keyword suggestion feature is likely to be an addition to the project keyword manager page. On entering the page, it will suggest keywords that can be quickly approved. We are hoping that this results in less time spent managing keywords and in turn drives data quality and improved search experiences. We are currently at work on this feature and hope it is something customers will be able to try in the near-term.
If this technique proves successful, we may try expanding the ontology to more detailed terms and develop a suggestion feature for file level keywords. As AI technology advances, it may become possible to fully automate the application of keywords. By remaining industry-focused and gradually incorporating greater levels of AI-assisted automation, we are ensuring that our product is delivering tangible benefits to users while preparing for the road ahead.