Picturepark Content Platform: Automation & Artificial Intelligence
By Picturepark Communication Team • Sep 21, 2017
Blog taken from a series of pages illustrating the functionalities of the Picturepark Content Platform.
While the amount of content created and used in more decentralized systems is growing, we are faced with limited resources in human labor for manually managing content. This directly raises the need to process content in more automated ways.
Through automation and artificial intelligence, large amounts of data can be processed many thousand times faster than via humans.
Through automation and artificial intelligence, large amounts of data can be processed many thousand times faster than via humans. This carries a huge potential for cost savings and faster time to market.
Additionally, some work previously executed partially only or not done at all (like tagging huge amounts of generated content) now gets finally done, increasing the general availability of content.
Samples of AI & Automation
- Auto-tagging of images, videos, documents and virtual content
- Transcribing audio and video, making it searchable
- Matching content with e.g. names of products or people
- Grouping or categorizing content based on similarities
- Connecting e.g. images with documents they are used in
- Detecting sensitive content (e.g. personal data) for approval
- Detecting duplicate content for easier cleanup
- Enriching content with additional contextual information
- Offer context when searching or tagging
- Suggesting related content based on searches
- Publishing of approved content, or archiving
- Processing or conversion of content based on activities
Limitations of Automation
However promising automation, Artificial Intelligence (AI) and Machine Learning (ML) generally are, nothing still outperforms the human brain when it comes to solving problems that are non-repetitive and require creative and “outside the box” thinking.
For example, auto-tagging content with generic keywords can provide very accurate matches for some categories of content such as stock images but it usually fails for more specialized content like particular products of a vendor, or content that requires context such as “company x CEO experimenting with technology when he was 10 years old”.
While it’s relatively easy to tag content accurately with a large amount of generic terms, matching these terms to a more relevant and specific vocabulary with fewer terms is a big challenge: There is a tower and a bridge on an image, but how does the algorithm know the tower is made of specialty steel offered by the company under SKU number 562789?
In general, wrong use of automation and artificial intelligence can lead to improperly tagged content and wrong decisions, and raise users distrust of your content collections.
We strongly believe that effective and sustainable automation goes beyond the buzz of Artificial Intelligence (AI), Machine Learning (ML) and a set of APIs. At the cornerstone of success stands an intelligently designed information, process and system architecture with well organized and highly-structured data. Then, artificial and human intelligence must be applied where each scales best.
As a sample, while content categorized as “stock content” is fine to be tagged automatically so users find their “blue sky” images, product-alike content better takes data from the filename or embedded metadata and relates this to existing product lists. This requires standardization of file name or XMP conventions up-stream the digital content supply chain. You can’t forego doing this when aiming full automation with zero manual interaction required further downstream.
Good use of automation limits human labor to a minimum e.g. by processing “big data” and filtering it for manual human verification prior to publishing. For example, images with faces might require you to request consent of the pictured persons. Or content that is auto-tagged needs to be verified by a human user ticking off the wrong associations and adding tags where needed.
Designed for Automation
The Picturepark Content Platform has been designed for automation and the use of Artificial Intelligence (AI) and Machine Learning (ML) technologies. This isn’t done via a single feature but a set of capabilities working together:
- Strictly structured approach to content and metadata
- Adaptive Metadata for relevantly described and controlled content
- Semantic relationships for context-rich information and guidance
- API-first design and microservice architecture for connectivity and interoperability
- Adherence and enforcement of standards, including customer-defined ones
- REST API and SDKs with Livestream and Service provider framework
The Picturepark Content Platform enables you to integrate any other system as needed, and have them work together in highly automated ways.
Picturepark also features ready-made workflows for auto-tagging and processing images, videos and documents, and off-the-shelf Connectors for data exchange with other systems or workflow engines.