High level overview

The process consists of two parallel workflows: one for candidates and another one for recruiters.

Preprocessing

Candidate processing is asynchronous and consists of the following steps:

• Information Extraction (This should be reviewed with recruiters experts to incorporate domain knowledge)
  • Extract job experiences/descriptions as individual sentences
  • Extract keywords for skills, technologies, abilities, and diplomas
• Embedding Generation
  • For each candidate, each extracted sentences are embedded and added to a vector database
• Hybrid Knowledge Graph Generation
  • Create “real” nodes for keyword matching
  • Create “virtual” nodes for semantic matching

Recruiter processing is synchronous and is composed of :

• Automatic job extraction
  • Skills / Previous job requirements / Diploma requirements
  • Job description for semantic search

graph LR
subgraph High_Level_Overview [High Level Overview]
    subgraph Recruiter
        subgraph Sync_Process [Synchronous Process]
            A[Recruiter] --|Automated Extraction|--> B(Job requirements)
            A --|Manual Modification|--> B
        end
    end

    subgraph Candidates
        subgraph Async_Process [Asynchronous Process]
            A1[Candidate] --> B2(Information Extraction)
            A2[Candidate] --> B2
            B2 --> C2(Knowledge Graph Generation)
            B2 --> D2(Embeddings generation)
        end
    end

    B --> A3(Recommendation System)
    D2 --> A3(Recommendation System)
    C2 --> A3(Recommendation System)
    A3 --> A4(Top K Candidates)
end

                                                   *Fig 1 : High level overview schema* 

Recommander system

It receives as input the Knowledge Graph, the vector database and the job requirements( mix of keywords and sentences)

• Look for exact matching in the knowledge graph for each keyword
  • Scoring function to rank candidates (integrate domain knowledge / user preference in the scoring i.e skills over diploma etc..)
• Semantic search to match job description and candidate’s previous experiences / description

Final scoring combining both scores

Add Clustering abilities to go beyond exact matching in the knowledge graph

Technical Needs Recommender System