Multi-mode matching
Strict, normalized and semantic matching for the same CV and ad, with the difference explained.
“Keyword match” opens after the analysis.
Three matching modes run over the same CV and ad:
- Strict: Exact, word-for-word matching.
- Normalized: Uses synonyms, abbreviations and a skill taxonomy.
- Semantic: Embedding-based matching that runs in your browser.
You see how each mode produces different results and which terms account for the difference. Semantic hits are shown as "possible match", never as confirmed.
How do the three matching modes differ in a resume keyword checker?
The resume keyword checker runs the same job description terms against your CV in three modes: strict, normalized and semantic. Each mode stands for a different kind of filter an employer might use.
Strict compares exact spellings only. Normalized adds a synonym table, abbreviations and, for Turkish and German, stemming and compound handling. Semantic compares meaning with a model that runs in your browser. The panel shows how many terms each mode matches and names the terms that explain the difference.
| Mode | How it matches | Example |
|---|---|---|
| Strict | Exact spelling, case ignored | The ad says Kubernetes, the CV says k8s: not matched |
| Normalized | Synonyms, abbreviations, stemming | k8s matches Kubernetes; js matches JavaScript |
| Semantic | Meaning, using a local model you turn on | A nearby passage is shown as a possible match |
Which mode sets the score?
Only the normalized mode sets the keyword score. Strict and semantic results are shown beside it as a comparison, so the score never moves with the mode.
The keyword dimension is worth 25 points: 15 for the terms listed as required and 10 for the rest. Terms under a requirements heading weigh 1.5 times, nice-to-have terms 0.8 times. Coverage of about 70 percent of the weighted terms earns a share in full. Terms that match only through a variant spelling, such as an abbreviation, cost one point, because a filter without a synonym table would miss them.
Example: what does the job description keyword match report?
For a job description keyword match, the panel lists every term from the ad as found, alias or missing, grouped as required, other or nice to have. It then counts the matches per mode.
Say the ad asks for Kubernetes and PostgreSQL and the CV says k8s and Postgres. Strict finds neither, normalized finds both and marks them as matched through an alias. The report says so, and suggests writing both forms once where they fit naturally, for example Kubernetes (k8s). Terms repeated more than 12 times are flagged as keyword stuffing.
What does semantic matching not do?
Semantic matching never confirms a match. It only marks a term the literal matcher missed as a possible match, with the closest CV passage and its similarity, and it never changes the score.
The model downloads only after you agree, runs entirely in your browser and sends no CV text anywhere. A hint appears only when the closest passage reaches a similarity of 55 percent. A term that is genuinely written in your CV is never downgraded to a guess. Treat a possible match as a prompt to check your wording, not as proof.
“Keyword match” opens after the analysis.
Last updated