Spaces:
Running
Running
add distance modifer to punish and award
Browse files- services/job_matcher.py +33 -1
services/job_matcher.py
CHANGED
|
@@ -47,6 +47,7 @@ class ScoreBreakdown(BaseModel):
|
|
| 47 |
category_score: float
|
| 48 |
seniority_score: float
|
| 49 |
experience_score: float
|
|
|
|
| 50 |
|
| 51 |
|
| 52 |
class MismatchAnalysis(BaseModel):
|
|
@@ -389,7 +390,7 @@ class JobMatcher:
|
|
| 389 |
experience = _experience_fit(cv_years, job.min_years_experience)
|
| 390 |
|
| 391 |
# ββ Weighted aggregation βββββββββββββββββββββββββββββββββββββββββ
|
| 392 |
-
|
| 393 |
0.42 * skill
|
| 394 |
+ 0.23 * semantic
|
| 395 |
+ 0.12 * title
|
|
@@ -398,6 +399,10 @@ class JobMatcher:
|
|
| 398 |
+ 0.05 * experience
|
| 399 |
) * 100
|
| 400 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 401 |
# Guardrails
|
| 402 |
if job_technical and not matched_technical:
|
| 403 |
final = min(final, 34.0)
|
|
@@ -442,6 +447,7 @@ class JobMatcher:
|
|
| 442 |
category_score=round(cat_score * 100, 2),
|
| 443 |
seniority_score=round(seniority * 100, 2),
|
| 444 |
experience_score=round(experience * 100, 2),
|
|
|
|
| 445 |
),
|
| 446 |
mismatch_analysis=MismatchAnalysis(
|
| 447 |
reason=reason,
|
|
@@ -508,3 +514,29 @@ class JobMatcher:
|
|
| 508 |
cv_vec = self._embed(f"job title: {cv_title}")
|
| 509 |
job_vec = self._embed(f"job title: {job_title}")
|
| 510 |
return max(0.2, min(1.0, _cosine_similarity(cv_vec, job_vec)))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
category_score: float
|
| 48 |
seniority_score: float
|
| 49 |
experience_score: float
|
| 50 |
+
category_distance_modifier: float = 0.0
|
| 51 |
|
| 52 |
|
| 53 |
class MismatchAnalysis(BaseModel):
|
|
|
|
| 390 |
experience = _experience_fit(cv_years, job.min_years_experience)
|
| 391 |
|
| 392 |
# ββ Weighted aggregation βββββββββββββββββββββββββββββββββββββββββ
|
| 393 |
+
base_final = (
|
| 394 |
0.42 * skill
|
| 395 |
+ 0.23 * semantic
|
| 396 |
+ 0.12 * title
|
|
|
|
| 399 |
+ 0.05 * experience
|
| 400 |
) * 100
|
| 401 |
|
| 402 |
+
# Category Distance Booster / Punisher
|
| 403 |
+
cat_modifier = self._category_distance_modifier(cv_category, job.category)
|
| 404 |
+
final = base_final + cat_modifier
|
| 405 |
+
|
| 406 |
# Guardrails
|
| 407 |
if job_technical and not matched_technical:
|
| 408 |
final = min(final, 34.0)
|
|
|
|
| 447 |
category_score=round(cat_score * 100, 2),
|
| 448 |
seniority_score=round(seniority * 100, 2),
|
| 449 |
experience_score=round(experience * 100, 2),
|
| 450 |
+
category_distance_modifier=round(cat_modifier, 2),
|
| 451 |
),
|
| 452 |
mismatch_analysis=MismatchAnalysis(
|
| 453 |
reason=reason,
|
|
|
|
| 514 |
cv_vec = self._embed(f"job title: {cv_title}")
|
| 515 |
job_vec = self._embed(f"job title: {job_title}")
|
| 516 |
return max(0.2, min(1.0, _cosine_similarity(cv_vec, job_vec)))
|
| 517 |
+
|
| 518 |
+
def _category_distance_modifier(self, cv_category: str, job_category: str) -> float:
|
| 519 |
+
"""
|
| 520 |
+
Calculate a score modifier between -20 and +20 based on how far apart
|
| 521 |
+
the CV category and Job category are semantically.
|
| 522 |
+
"""
|
| 523 |
+
if not cv_category or not job_category:
|
| 524 |
+
return 0.0
|
| 525 |
+
|
| 526 |
+
cv_cat_clean = cv_category.strip().lower()
|
| 527 |
+
job_cat_clean = job_category.strip().lower()
|
| 528 |
+
|
| 529 |
+
if cv_cat_clean == job_cat_clean:
|
| 530 |
+
return 20.0
|
| 531 |
+
|
| 532 |
+
cv_vec = self._embed(f"industry sector: {cv_cat_clean}")
|
| 533 |
+
job_vec = self._embed(f"industry sector: {job_cat_clean}")
|
| 534 |
+
|
| 535 |
+
similarity = _cosine_similarity(cv_vec, job_vec)
|
| 536 |
+
|
| 537 |
+
# Map similarity [0, 1] to [-20, 20]
|
| 538 |
+
# Assuming similarity < 0.5 is distant (-20 to 0)
|
| 539 |
+
# Assuming similarity > 0.5 is close (0 to 20)
|
| 540 |
+
modifier = (similarity - 0.5) * 40.0
|
| 541 |
+
|
| 542 |
+
return max(-20.0, min(20.0, modifier))
|