If it’s a job title, a lot. We built a tool that plumbs the deep reservoir of skills connected to thousands of job titles.
What skills does a particular job require? What skills are in demand?
To answer these questions, ADP set out to design a robust tool to examine the relationships between jobs and skills. We identified and assembled more than 100,000 unique proficiencies and nearly 9,000 job titles, which we used to create a skills ontology graph that identifies how jobs and skills are related, and a whole lot more.
Our skills ontology is rich with potential. In fact, it’s already delivered a major finding: Being a responsible adult who can make good decisions is the most in-demand collection of skills, regardless of the job. Read on for details.
What’s an ontology?
A graph is a mathematical structure used to model relationships between objects or entities. It consists of nodes (the objects or entities) and edges (the relationships). A taxonomy is a classification system typically organized according to a schema or hierarchy of nodes.
An ontology is built using these graphs. It’s a data model that shows the relationships between things. It’s often drawn as a collection of dots which represent the entities or objects you’re interested in. Lines connect the dots that have a relationship to one another.
Table 1: A diagram depicting the basic structure of the ADP skills ontology, focusing on one particular job. Not all skills or jobs are depicted.
The ontology adds a semantic layer to the relationships, so the relational edges have some human-understandable meaning. In other words, the semantic layer in the ontology allows people to identify and make sense of these relationships.
In the ADP skills ontology, the nodes represent one of two kinds of objects or entities: jobs and skills. To build the ontology, we first had to assemble taxonomies for each.
What’s a taxonomy?
A taxonomy is a system of classification of entities, typically organized according to some schema or hierarchy.
Our ontology has two entities: jobs and skills. We add the relation (edge) that asserts that a job has some set of skills.
The job taxonomy is organized hierarchically by job function and sub-function. For each job, a generative AI model was used to produce a hierarchically organized JSON object that relates specific, granular skills to their higher-order “parent” skill. What we’re left with is a richer way of viewing relationships between jobs and skills.
The skills also are organized hierarchically due to the output of the generative AI, asserted by the structure provided in the prompt.
What’s a skill?
A skill is a particular ability, or body of experience or knowledge. A particular skill can be associated with more than one job. Some jobs require dozens of skills.
“Juggling” is a skill associated with jugglers, clowns, and acrobats. “Animal handling” is required of elephant keepers, dog groomers, and zookeepers; using spreadsheets is a must for accountants, but also for store planners and even herbarium technicians.
Skills also have an associated function and subfunction, which allows the ontology’s users to differentiate between different applications of the same skill. Police detectives and used-car salespeople both utilize the skill of “negotiation”, but one would expect these two groups to apply their expertise in different ways.
To differentiate, the job of used-car salesperson pairs “negotiation” with the function “sales and marketing” and the subfunction “field sales”. For police detectives, “negotiation” is under “public safety” and “law enforcement”.
So. How does it all work?
By looking at the words that describe a skill, our ontology allows us to explore the relationship between the skill and its function. Using latent semantic analysis, as described in the methodology section below, we can identify words that are strongly associated with particular skill functions
A skill with the words “selling”, “franchise”, or “channel” in the name often is associated with jobs in sales and marketing, retail trade, and customer service.
If you’d describe a skill using the words “art”, “adobe”, or “color”, you probably are talking about a skill used in occupations in entertainment, creative, or specialized design services.
Skills containing the words “document”, “case”, or “courtroom” are found in business, legal, public safety, financial services, or government.
And skills containing the words “animal”, “event”, or “fitness” frequently are found in jobs with diverse functions in science and mathematics, business, entertainment, public safety, transportation, storage and logistics, manufacturing, environment, and travel.
Being in shape and possessing animal-focused skills clearly opens doors.
When we populate the ontological structure with data from our job and skill taxonomies, we get a graph that identifies whether or how jobs and skills are related. That is, it shows which skills are associated with which jobs, and vice versa.
The ADP skills ontology further splits skills into two groups: general skills and unique skills. General skills can apply to any number of jobs, while unique skills tend to be more specialized to a particular industry or job type. Of our over 100,000 skills, nearly 17,000 are general skills commonly found in many jobs and industries, and nearly 98,000 are unique skills which tend to be limited to specific industries and jobs.
For example, under “health care and social assistance” is a subfunction, “nursing services”. Within “nursing services” is a list of skills: monitoring vital signs, life support, patient care, mobility assistance, anatomy, physiology, and so on, for 12,555 words. That’s a big skill set.
What did we find?
Our first foray into the ADP skills ontology found that the most common general skills across the nearly 9,000 jobs covered by the ontology are:
- Communication
- Problem-solving
- Time management
- Written communication
- Verbal communication
These can be somewhat glibly summarized as “the ability to behave like a responsible adult.”
The most common unique skills in the ontology are:
- Regulatory compliance
- Workplace safety
- Quality control
- Risk management
- Inventory management
These can be summed up as “the ability to make good decisions.”
In short, being a responsible adult who can make good decisions is the most in-demand collection of skills across all jobs, according to the ADP skills ontology.
What does the future hold?
The ADP ontology is uniquely rich and granular because it’s linked to the ADP payroll data ecosystem. And by using ChatGPT and the billions of sources it accesses, the ADP ontology achieves massive scale relative to skills ontologies that are derived mainly from social media platforms that host resumes and job listings.
This enormous and broad source data—called a corpus in natural language processing lingo—makes the ADP ontology less susceptible to distortions, such as those that can result from applicants whose resumes are written to game screening algorithms, or from employers whose job descriptions are written to attract applicants.
The ADP skills ontology will be put to good use. We can tap it to spot highly valued or in-demand skills, understand similarities between different job tracks, or identify skills that are commonly associated with each other. Watch for articles that leverage this new ontology.
Methodology
To create the association between words used to describe a skill and the function of that skill, we used a latent semantic analysis. We treated each subfunction within a function as a document and concatenated the names of all the skills associated with that subfunction to be the content of that document. Words that made up more than 25 percent of a document were removed. From this we created a document-term matrix with 11,620 unique words. A singular value decomposition (SVD) was performed on the document-term matrix to reduce the dimensionality of the documents from 11,620 to 100, maintaining about two-thirds of the variance. Finally, a k-means clustering was applied to the reduced matrix to cluster the documents into function groups using the skill words in the documents.
To learn more about our skills ontology and its methodology, email [email protected].
Nina Plotko contributed to this report. Nina Plotko is a senior data scientist on ADP’s OneAI Team, studying payroll and HR data in the ADP ecosystem. She helped develop the skills ontology and improved the classification of individuals into the job taxonomy.
