only a minor amount of development effort by
reusing existing code from the skill extraction
engine. In fact, mappings from normalized skills to
their associated skill lists were already a feature of our
data set. This new functionality was simply a matter
of ( 1) normalizing the entered skill, and ( 2) returning
an ordered list of its associated skills. Today, this feature adds customer value by enabling semantic
expansions of candidate search queries written by
recruiters, and by providing valuable data to hiring
managers during the prehire process. Applications of
this feature include offering career-path suggestions
to job seekers and aiding job posters with fine-tuning
description sections in job postings.
Large-scale adoption of the service has resulted in
many improvements and enhancements to the system over time. WSD was added to the service in version 2.0 after users requested a more robust skill disambiguation mechanism. Continuous feedback
loops from users also helped in removing noise from
the taxonomy. Skill categorization was added to the
taxonomy after users suggested that there was a market need for it in several products. On the user-facing
side, we are also working to give our users the option
to see the skills extracted from their resume the
instant they upload their resume on the site. This feature will also allow users to make corrections or additions to their list of skills, and we will be able to track
these edits to further improve our understanding of
how our users reason about skills.
As part of operating and maintaining the SKILL
Conclusion and Future Work
The skills gap is increasingly cited as the main reason
for the disparity in a large number of job openings
and underemployment in economies around the
world. We argue that accurately detecting skills in
human capital data is a first step to resolving this
Figure 5. Skills Traffic Over a Three-Day Window.
Limited to one single caller who executes requests in massive batches. The y-axis indicates total calls per 30 minutes.
Sat Jan 20 12 PM Sun Jan 21 12 PM Mon Jan 22