Skip to content

Latest commit

 

History

History
63 lines (48 loc) · 3.25 KB

matching.md

File metadata and controls

63 lines (48 loc) · 3.25 KB
Raw

Matching algorithm idea description

Elements:

1.Skills + Experience :

assumed predefined experience levels (beginner, intermediate, advanced), and predefined skills set

  • the sum of values for every skill of job/candidate:
    • if the skill does not match - 0
    • if the skill matches and the level is the same or the candidate level is higher than required in the job -> 1
    • if the candidate level is lower than job required - if combination (cand: beginner job: intermediate) or (cand:intermediate job:advanced) -> 0.7
    • if the cand:beg job:adv -> 0.3

2.Spoken Languages:

assumed predefined language levels (beginner, intermediate, advanced), and predefined language set

  • the sum of values for every language of job/candidate:
    • if the language does not match - 0
    • if the language matches and the level is the same or the candidate level is higher than required in the job -> 1
    • if the candidate level is lower than job required - if combination (cand: beginner job:intermediate) or (cand:intermediate job:advanced) -> 0.7
    • if the cand:beg job:adv -> 0.3

3.Location:

(assuming that locations are predefined cities with their coordinates that enable distance count)

  • measure the distance between the seek locations from the candidate and offered locations in the job and average them

4.Job location type:

(remote / office/ hybrid)

  • value depends on combinations found in the job and candidate preferences:
    • if candidate and job contain the same value -> 1
    • if existing combinations=: remote/hybrid or hybrid/office -> 0.5
    • if no match( combination remote from one side and office from another)-> 0

5.Employment type:

(internship or long-term)

  • if match is found in employment types preferred by candidate and offered by job -> 1

6.Employment rate:

(full-time or part-time)

  • if a match found in employment rate types preferred by the candidate and offered by job -> 1

7.Values:

  • language model for finding value entities in both the values of the candidate and company(that posted a job) and calculating how many matched -> sum of matched

8.Job description fields / Candidate about me field:

  • language model to compare similarity of candidate 'About me' field and combined job description field on the basis of recognised entities presence and synonyms match, sentiment -> similarity output value

Priority and weights:

  1. Skills Experience -> 1
  2. Spoken Languages -> 0.7
  3. Location distance -> 0.6
  4. Job Location type, Employment type, Employment rate -> 0.5
  5. Values, Job description fields / Candidate about me field -> 0.4

Algorithm:

  • Build vector having the elements on subsequent positions for every item of the matched entity (job for candidates and candidates for jobs)
  • Apply weighting to positions
  • Calculate distances between the base vector and the matched entity vectors and choose n closest ones and return them. (Due to the fact that different positions of the vectors will have different value scales (some numerical and some categorical) - the matching algorithm should take into account this property)

Companies - candidates match can be done in the same way, having all jobs posted by the company as company representation and then calculating the summed distance of vectors