(1) Extracting the keys of the object is easy.
(2) Mapping the keys to a ontology predicate is easy for small numbers of keys.
(3) Extracting relevant keywords from the job descriptions is harder
(for 3) At a crude level mapping the words to keywords could be done through tokenization and manual mapping. But what about something fancier like Apache Stanbol? https://stanbol.apache.org/docs/trunk/components/enhancer/nlp/