How Does the Career Site Job Recommendation Feature Work?

[Christopher_Nickel-Felton]

You’ve probably noticed that SuccessFactors Recruiting is starting to catch up around skills. Our first skills feature we delivered was Assisted Screening for the new Applicant Workbench. Our new skills feature is Job Recommendations for candidates on your career site. Candidates can upload their resume, get job recommendations, and then see how their skill set matches to a job.

We’ve got a great home page component for getting the candidates attention and building engagement with the discovery process. Its highly configurable and supports multiple layouts. You can use it on any page, but your homepage is probably a good place to start.

After parsing the Candidates resume for skills we show the top matching jobs. All the filters can be used in combination with the skill-based recommendations. The job map works with skills too!

When the candidate clicks into a recommended job, they will see how their skill set relates to that job posting.

It’s very straight forward for the Candidate, but I’m starting the hear customer questions such as:

• What happens to the resume?
• Are there data privacy concerns here?
• Do I need to set up all my skills in Talent Intelligence Hub first?
• What is this actually doing, how does it work?

So, what happens to the resume?

The resume is examined for skills and only skills, no personal information is parsed or extracted, no images are parsed. We support text, pdf and Word documents. Once the skills are identified the resume is immediately discarded. Everything in this process is anonymous as much as possible, no personal data is used and saved in any way.

Are there data privacy concerns here?

SAP AI Ethics, SAP Data Privacy Team, and SAP Legal, were all engaged in the development of this feature, all applicable SAP standards were adhered to. You can review these standards in the SAP Trust Center. Even though the resume is processed anonymously, resumes do contain personal information. For this reason, we included the capability to present the candidate with a consent statement, which they must accept, to use this feature. Presenting this consent statement is optional. Your legal team may not feel this is necessary, but it is available for you to use.

Do I need to set up all my skills in Talent Intelligence Hub first?

No. This feature uses our Universal Ontology of Market Skills and does not require you to set up your own skills library.  The Universal Ontology represents the generally understood skills and professions that are in use in human work efforts across the world.

If you think about it, this makes sense. The external job seekers most likely do not know the inner workings of your organization and may not relate their previous work experiences and abilities to your bespoke skills model defined in Talent Intelligence Hub, they have crafted their resume in relation to the entire job market.

What is this actually doing, how does it work?

This is the technical part of this blog, so you can stop now if you’re not interested in seeing behind the curtain. There’s no statistical math involved, this is a layman’s explanation, well, there is a little math at the end…

At a high level, Job Recommendation involves 3 steps:

1. Identifying the skills which are required for a given Job Requisition:
         Through machine analysis - Automated Skill Determination
         Through human refinement and validation - Manual Skill Refinement
2. Identifying the skills which the prospective applicant possesses
3. Evaluating the intersection of these two sets of skills to calculate which Job Requisitions are a good fit and should be recommended

Automated Skill Determination

 The job posting title and description are related to the Universal Ontology. Two sets of skills are identified in this process. First, skills which are directly correlated to the job are identified, these the Job Skills. Second, the job is related to a list of Standardized Professions in the Ontology. The critical skills required for these professions are then identified. These are the Profession Skills. 

The overlap of the Job Skills and Professions skills is used to determine which skills the automated process considers to be primary skills. Job Skills which have no overlap with Profession Skills and labeled as secondary skills. Profession Skills with no direct match to the Job are then discarded.

Manual Skill Refinement

At this point the Job Postings are ready to use on the Career Site for Recommendations. The automated skills can be reviewed and manually adjusted as needed from the Requisition detail screens. Skills can be set as primary or secondary. The Universal Ontology can be searched for any additional required skills that were not automatically identified.

Resume Analysis

The Career Site visitor uploads a resume, which is analyzed and related to the Universal Ontology. A Resume Skill Profile is created from this process.

How Job Recommendations Are Calculated

The Resume Skill Profile is combined with any additional user entered search criteria to calculate a Recommendation Score. The Job Recommendations are returned in order of this score.

Primary skill matches are scored higher than secondary skill matches.

How frequently a skill or keyword appears in the job postings also affects the score. Matches to skills or keywords which appear less frequently in the set of job postings are scored higher than skills or keywords which appear more frequently.  

Recommendation Quality Checks

To reduce the chance of low-quality recommendations, especially when a resume does not strongly relate to the available job postings, two additional quality checks are run.

The Recommendation Scores for each potential recommendation are compared. If a significant drop off in score is detected within the recommendation list, then jobs below this threshold are discarded.

Finally, the percentage of skills matched is examined for each potential recommendation. If 10% of primary skills, or 25% of secondary skills do not match, then the recommendation is discarded. No magic, just math.

Last Notes

The Career Site Job Recommendations feature requires Career Site Builder on Unified Data Model (UDM). If you’re not on UDM yet, you should take a look, we’re delivering a lot of great features on this new platform, Job Recommendations is just one of many to come.