Nowadays, everybody is familiar with the idea of search. We type a query into a search engine and get back an ordered list of relevant documents. There are many ways to achieve such a result but the majority of approaches employ some form of scoring function that is used to rank the set of documents based on their relevance to the query.
Recommending candidates for a vacancy is no different: we begin with a query – the vacancy – and return the most suitable candidates from the pool of applicants by looking at things like skills the candidates have, and the companies that they have worked at. This is great if all that you want to do is find the best candidate from a pool of applicants. Unfortunately, it won’t usually tell you whether the best candidate is actually any good.
As a rule the scoring function is not easily interpretable since it is formed by a number of interacting influences. For example, say our scoring function measured the length of time that each of the candidates has used a given skill over the last 10 years. We know the maximum score achievable and so it should be straightforward to determine how good each of the candidates is in some absolute sense. If the best candidate has only three years of experience we know that we probably need to re-advertise or offer more money. So far, so simple.
Things get trickier when we consider the possibility that no-one has more than three years’ experience of the skill because it is a new technology. Using an absolute score we would conclude that there were no good candidates, even if the most suitable person on the planet had applied. Similar things can happen if an unrealistic combination of requirements are requested in the vacancy. If we search for a ‘unicorn’, then none of the candidates that apply for a role will get high scores for every requirement and so we must conclude that all of the applicants are below par in some way.
Elevate’s approach: market scoring
To address this problem and provide recruiters using Elevate with a more interpretable set of results we have developed the idea of a market score. Market scoring is simple but powerful. Instead of simply ranking the applicants to a given vacancy, we search our database of resumes and create a population of people who would be suitable for the role. These are people who have the right sort of skills and experience but for whatever reason have not applied to the vacancy. This set of profiles represent the ‘market’ and reflect the availability of different skill sets in the wider population. We add the real applicants to this population and rank the whole set.
Now, if one of the applicants comes top of the rankings we can say with far more certainty that they are a good candidate and should be considered for shortlisting. Conversely, if all of the applicants are ranked mid to low relative to the market population then we can identify that the best applicant is not good enough in some objective sense and we should re-advertise or offer more money. The approach elegantly handles unusual sets of requirements as well as the problems outlined above and allows recruiters to make informed judgements and quickly ascertain the quality of applicants to a given vacancy.
Written by Bart Baddeley, Elevate Direct’s Chief Data Scientist. Bart has 20 years’ experience in the fields of artificial intelligence and machine learning. He has a 1st class honours degree in artificial intelligence and neuroscience and a PhD in neural network modelling.
If you’d like to discover more about Elevate Direct’s AI platform and how it can help you, please get in touch.
Share this article in your social networks.