At Feathercap we’ve doubled down on our belief that no one wants to curate and manage content. Introducing unsupervised curation in Feathercap
In our previous post: “5 reasons why learning content curation is getting difficult” we talked about how its getting too hard for existing systems like LMSs and many of the new learning experience systems (LXPs) to find the answer to a specific job question. These vendors to date have a noble goal; providing a curriculum of content to employees to move them toward their personal learning goals as well as in their career. The trouble is, there is not enough emphasis on answering specific employee questions that they need to do their job. These systems are focused on organizing a curriculum either mostly by hand or with some delving into auto tagging content using AI with the hopes of removing the need to manually create general topic maps of all their content. This approach though helpful in generally categorizing content still doesn’t deliver the exact answer and page from the question.
An example skill we’ve run into: “How do I set-up an outside patio table?” - a bus person at a restaurant.
The answer to this doesn’t require a curriculum, or a course. If we know where the bus person is working and if we have access to the restaurant and all of their content (PDFs, PPTs, Word docs, emails, images saved, etc) in all locations we then can curate it all. But how do we arrive at the answer to this question? The best result: the bus person just needs to see an image of a correctly set up patio table. It's obvious as we discuss it here, but when 100s, 1000s or more content must be examined for the answer to be found, its not easy and predictably get the exact required version of content at exactly the right time.
This leads to the second reason we incorporated machine learning/ NLP into Feathercap.
Getting the exact right answer from an employee question. Why we added semantic driven search.
We’ve mentioned the value of auto tagging content, this makes it easier for the right content topics to be created and have everything align to them. That’s a great start and we use advanced LDA to perform this, which is an unsupervised approach to curation that lets the system "self-learn" it own categorization. But to be able to semantically understand what a question means and then scan through all available content, anywhere in the organization requires successfully deploying additional AI, specifically machine learning and natural language processing. This is how successful on-the-fly language translation is done. To be successful it’s necessary to understand the meaning of words, phrases, entities (person, place, thing), sentiment as well as syntax. Because of all this we are able to pull out what a specific paragraph, image, video, page or entire set of pages comprising content semantically means along with the full meaning/ sentiment and goal of any employee question and then match them. Coincidentally, because of the above we can work with questions and content in multiple languages.
So now when you use Feathercap, all of your PDFs, PPTs, videos, images, word docs and most any other content is automatically turned into trackable learning content and answers. (See our post: “Instantly turning all your companies knowledge into learning content has arrived”) By also entering a question, phrase, word pairing of any text we can pull the exact answer to your question along with the associated paragraph and page it came from. We also automatically track the employees experience with the content to further help decide its usefulness to them and others.
See our AI and the Augmented Workforce primer on how a workforce and technology can effectively work on tasks together.