For automated matching between jobs and candidates, skill-recognition (in resumes and Job-ads) is essential for making the best possible matches. Over the last few years, Textkernel worked on automated skill-extraction and -normalisation for many languages. The technology is starting to mature and it is finding its way to our customers and into our products.
Bauke Visser, Textkernel’s Data Consultant, tells about his Innovation Week project in which he 3D representation of these skills.
By Bauke Visser
There are many ways ways to analyse these thousands of skills in a way that brings value to customers. Possible areas of interest might be:
- Relatedness between skills (which skills occur together frequently)
- Identification of skills that are most valuable for a certain profession
- Identification of skills that are emerging and which are disappearing
Relatedness between skills can also help when grouping skills together in a classification. For the English language, we currently identify approx. 12.000 unique (normalised) skills. We recognise these from many more spelling-variants and synonyms.
Textkernel Innovation Week 2019 – The 3D Skill Explorer
As part of Textkernel’s annual Innovation Week, we created a 3-dimensional visualisation of these 12.000 skills. It represents the skills in a way that relatedness in the job-market results in closeness in the visualization.
This innovation is particularly unique because it offers professionals interested in the evolving skill landscape an immediate and visual manner to explore the skill landscape. Recruiters, business strategists and business owners can now get a very in-depth understanding of how skills are interrelated and evolving over time.
Viewers will gain an atmospheric experience within the viewer that provides a constellation of skills. Each dot in a cloud represents one of the 12.000 skills that we extracted from 14 million jobs (this is after deduplication; so the original dataset contains ~100 million job-ads).
The closer the dots are together in the cloud, the more related they are in the labour-market (based on co-occurrence; skills appearing together). The colors represent the frequency; the red skills were mentioned relatively often, and the blue skills less often.
What we did
Information-extraction and -normalisation is at the core of what Textkernel does. We have an internal skill-extraction-service that can process documents with very high throughput. For this project, we extracted the skills from 14 million jobs, along with some essential metadata. To go from a table of 12.000 skills by 14 million jobs, to a list of xyz-coordinates, we used dimensionality-reduction methods called PCA and t-SNE.
We were able to make a VR-version of the application as well. The experience surpassed our expectations. One of the critical features is that the image reacts to movements of the head, so the user is able to look around in the skill-space. We were also able to add some basic controls: push a button to control the speed. This video gives an idea of the VR-experience (you will need a VR-viewer for it).
For VR-aficionados, we’d love to share the experience with you.
Interested? Please contact us for more information. We’d love to hear your feedback and your ideas about possible business-applications:
About Bauke Visser:
Bauke has been working as a Data Consultant with Textkernel since 2013. He has a passion for (Labour Market) statistics, complex quantitative problems and demanding customers.