API Technical Specs
This documentation is for Version 10 of the Sovren REST API, released on December 15, 2020. Both V9 and V10 use the same parsing and matching engines under-the-hood, but V10 is more streamlined and has a vastly simpler output. Please visit this link for an in-depth comparison.
This page is here to tell you how our AI Matching engine works, but most important is that it is different than every other engine out there. You have never worked with an engine that works this way, so it's very important to understand its differentiators.
Thinks like a human
Our AI Matching Engine thinks like a human. Matching starts by analyzing both documents for data points to determine what data is relevant and how important it should be. It also knows to count data that is more recent as more important and not to just do a naive keyword search. For example, if a recruiter is looking for a Java developer, a candidate that is currently a Java developer is a far better fit than a candidate that was a Java developer 5 years ago. This is exactly how a human would analyze a job to a stack of resumes, and also how our engine thinks.
Our AI Matching Engine is the only truly controllable engine on the market. Matching starts by analyzing the input document to decide what criteria are important. It then generates complex queries across 8 different categories and calculates a weight, or level of importance, for each category. Our engine will decide on weights by default, but in some cases, you will disagree and need to tweak the weights to influence the calculation. Optionally, you can specify your own weights on the initial match if you already know what is important.
It is crucial for recruiters to understand the 'why' behind the results they see. Distrust in a platform hinders user adoption. We built our engine to output all the information we considered to paint a clear picture of how we scored the transaction. We return an overall score, directional scores, and even scores broken down by category. We took it a step further and output terms found, not found, and written explanations for each of the terms considered in the query. We believe matching should be explainable, not magic.
When a recruiter is finding candidates for a job, they not only check if the candidate can do the job (one direction), they will also ensure the job is a good fit for the candidate (other direction). Most matching engines check a single direction, but ours checks both and outputs a unified bidirectional score called SovScore. For example, a recruiter has a poorly written job description that has no real discernable skills other than Microsoft Office. An office administrator that possesses that skill set (and much more) would be a good fit to that position, but from the candidate's perspective it is a lousy fit because the job would only utilize one skill of the many the candidate possesses; therefore, the candidate is very overqualified. Rather than pollute the results set with candidates who are capable but not interested, we evaluate each match from both perspectives.
Your task isn't simply to integrate a couple of API calls, it's to integrate ALL the differentiators. If you just use this for matching without these features, you will have hidden the true power and you users won't like the engine. Our engine stands out because it's controllable and understandable, and users need to be presented with this information to properly interact with the data.