Semantic job matching is a method of comparing a CV with a job description by meaning rather than by matching keywords. It finds a person who has done the work even when their job title and their words are different from the job's. "Semantic" means "to do with meaning". It is the matching engine inside most AI recruiting tools.
Keyword search fails in two ways. It misses good people who describe their work in other words. And it ranks people highly just because their CV repeats the right words. Job-seekers know this, so many of them fill their CVs with keywords, sometimes with help from a resume builder.
Semantic matching works differently. A language model reads the job description and turns it into a list of numbers, such as 768 numbers. This list is called a vector, or an embedding. The model is trained so that texts with similar meanings get similar lists of numbers. The page on vector embeddings explains how. The model turns each CV into a vector in the same way. The software then measures how close each CV's vector is to the job's vector. The closeness is a score, usually between 0 and 1.
Imagine a retail company in Kolkata that is hiring a data engineer. The job asks for someone who has built data pipelines in Spark. A data pipeline is a program that moves data from one system to another on a schedule, such as every night, and cleans it on the way. One CV comes from a person whose title is "MIS executive" at a consumer goods distributor in Indore. MIS means management information systems, the team that prepares reports. The CV never uses the words "data engineer". But it says that he wrote nightly PySpark jobs that load sales data from the billing system into the company's data warehouse. PySpark is the Python version of Spark. The semantic score for this CV is 0.86. A second CV has the title "data engineer", but it describes only weekly reports made in Excel. Its score is 0.61. So the first CV is ranked higher.
A good matching engine also shows the recruiter why it ranked a CV highly. In this case, it highlights the line about the nightly PySpark jobs. The recruiter still decides whom to call. The applicant tracking system stores the match score with the candidate.
The same method is used inside companies. An internal talent marketplace matches employees whose project has ended to new projects, by comparing what they have done with what each project needs.