Spaces:
Running
Running
improve calculator
Browse files- services/job_matcher.py +17 -53
services/job_matcher.py
CHANGED
|
@@ -11,15 +11,7 @@ from services.taxonomy_client import (
|
|
| 11 |
DEFAULT_SKILL_ALIASES,
|
| 12 |
DEFAULT_CATEGORY_SKILLS,
|
| 13 |
SENIORITY_ORDER,
|
| 14 |
-
_extract_skills,
|
| 15 |
-
_infer_category,
|
| 16 |
-
_extract_years,
|
| 17 |
-
_detect_seniority,
|
| 18 |
-
refine_seniority_with_years,
|
| 19 |
-
_extract_cv_title,
|
| 20 |
-
compute_years_from_experience,
|
| 21 |
)
|
| 22 |
-
from services.cv_chunker import _parse_cv_sections, _parse_experience
|
| 23 |
|
| 24 |
|
| 25 |
# ---------------------------------------------------------------------------
|
|
@@ -215,25 +207,26 @@ class JobMatcher:
|
|
| 215 |
self,
|
| 216 |
job_json: Dict[str, Any],
|
| 217 |
cv_markdown: Optional[str] = None,
|
| 218 |
-
section_embeddings: Optional[Dict[str, Any]] = None,
|
| 219 |
cv_title: Optional[str] = None,
|
| 220 |
cv_years: Optional[int] = None,
|
| 221 |
cv_seniority: Optional[str] = None,
|
| 222 |
-
|
|
|
|
|
|
|
| 223 |
cv_category: Optional[str] = None,
|
| 224 |
) -> Dict[str, Any]:
|
| 225 |
"""
|
| 226 |
Predict the match between one job and one CV.
|
| 227 |
-
Pass ``section_embeddings`` from ``/process-cv`` to skip re-encoding.
|
| 228 |
"""
|
| 229 |
cv_data = self._parse_cv(
|
| 230 |
-
cv_markdown=cv_markdown,
|
| 231 |
-
|
| 232 |
-
|
| 233 |
-
|
| 234 |
-
|
| 235 |
-
|
| 236 |
-
|
|
|
|
| 237 |
)
|
| 238 |
return self._predict_single(job_json, cv_data)
|
| 239 |
|
|
@@ -366,7 +359,7 @@ class JobMatcher:
|
|
| 366 |
semantic_matches = self._semantic_skill_matches(
|
| 367 |
missing_technical,
|
| 368 |
individual_skills_vecs,
|
| 369 |
-
threshold=0.
|
| 370 |
)
|
| 371 |
if semantic_matches:
|
| 372 |
matched_technical = sorted(set(matched_technical) | semantic_matches)
|
|
@@ -377,13 +370,13 @@ class JobMatcher:
|
|
| 377 |
|
| 378 |
extra = sorted(cv_skills - job_technical - job_soft)[:8]
|
| 379 |
|
| 380 |
-
tech_denom = min(len(job_technical),
|
| 381 |
-
soft_denom = min(len(job_soft),
|
| 382 |
-
|
| 383 |
tech_score = min(1.0, len(matched_technical) / tech_denom) if job_technical else 0.50
|
| 384 |
soft_score = min(1.0, len(matched_soft) / soft_denom) if job_soft else 0.50
|
| 385 |
|
| 386 |
-
skill = min(1.0, (tech_score * 0.
|
| 387 |
|
| 388 |
# ββ Component scores βββββββββββββββββββββββββββββββββββββββββββββ
|
| 389 |
cv_cat_lower = cv_category.strip().lower() if cv_category else ""
|
|
@@ -399,12 +392,9 @@ class JobMatcher:
|
|
| 399 |
experience = _experience_fit(cv_years, job.min_years_experience)
|
| 400 |
|
| 401 |
# ββ Weighted aggregation βββββββββββββββββββββββββββββββββββββββββ
|
| 402 |
-
# Discount semantic score by category score (prevents broad semantic clustering across domains)
|
| 403 |
-
semantic_weighted = semantic * cat_score
|
| 404 |
-
|
| 405 |
base_final = (
|
| 406 |
0.42 * skill
|
| 407 |
-
+ 0.23 *
|
| 408 |
+ 0.12 * title
|
| 409 |
+ 0.11 * cat_score
|
| 410 |
+ 0.07 * seniority
|
|
@@ -540,29 +530,3 @@ class JobMatcher:
|
|
| 540 |
cv_vec = self._embed_title(cv_title)
|
| 541 |
job_vec = self._embed_title(job_title)
|
| 542 |
return max(0.2, min(1.0, _cosine_similarity(cv_vec, job_vec)))
|
| 543 |
-
|
| 544 |
-
def _category_distance_modifier(self, cv_category: str, job_category: str) -> float:
|
| 545 |
-
"""
|
| 546 |
-
Calculate a score modifier between -10 and +10 based on how far apart
|
| 547 |
-
the CV category and Job category are semantically.
|
| 548 |
-
"""
|
| 549 |
-
if not cv_category or not job_category:
|
| 550 |
-
return 0.0
|
| 551 |
-
|
| 552 |
-
cv_cat_clean = cv_category.strip().lower()
|
| 553 |
-
job_cat_clean = job_category.strip().lower()
|
| 554 |
-
|
| 555 |
-
if cv_cat_clean == job_cat_clean:
|
| 556 |
-
return 10.0
|
| 557 |
-
|
| 558 |
-
cv_vec = self._embed(f"industry sector: {cv_cat_clean}")
|
| 559 |
-
job_vec = self._embed(f"industry sector: {job_cat_clean}")
|
| 560 |
-
|
| 561 |
-
similarity = _cosine_similarity(cv_vec, job_vec)
|
| 562 |
-
|
| 563 |
-
# Map similarity [0, 1] to [-10, 10]
|
| 564 |
-
# Assuming similarity < 0.5 is distant (-10 to 0)
|
| 565 |
-
# Assuming similarity > 0.5 is close (0 to 10)
|
| 566 |
-
modifier = (similarity - 0.5) * 20.0
|
| 567 |
-
|
| 568 |
-
return max(-10.0, min(10.0, modifier))
|
|
|
|
| 11 |
DEFAULT_SKILL_ALIASES,
|
| 12 |
DEFAULT_CATEGORY_SKILLS,
|
| 13 |
SENIORITY_ORDER,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
)
|
|
|
|
| 15 |
|
| 16 |
|
| 17 |
# ---------------------------------------------------------------------------
|
|
|
|
| 207 |
self,
|
| 208 |
job_json: Dict[str, Any],
|
| 209 |
cv_markdown: Optional[str] = None,
|
|
|
|
| 210 |
cv_title: Optional[str] = None,
|
| 211 |
cv_years: Optional[int] = None,
|
| 212 |
cv_seniority: Optional[str] = None,
|
| 213 |
+
cv_skills_canonical: Optional[List[str]] = None,
|
| 214 |
+
cv_skills_technical: Optional[List[str]] = None,
|
| 215 |
+
cv_skills_soft: Optional[List[str]] = None,
|
| 216 |
cv_category: Optional[str] = None,
|
| 217 |
) -> Dict[str, Any]:
|
| 218 |
"""
|
| 219 |
Predict the match between one job and one CV.
|
|
|
|
| 220 |
"""
|
| 221 |
cv_data = self._parse_cv(
|
| 222 |
+
cv_markdown=cv_markdown or "",
|
| 223 |
+
cv_title=cv_title or "Unknown",
|
| 224 |
+
cv_years=cv_years or 0,
|
| 225 |
+
cv_seniority=cv_seniority or "mid",
|
| 226 |
+
cv_skills_canonical=cv_skills_canonical or [],
|
| 227 |
+
cv_skills_technical=cv_skills_technical or [],
|
| 228 |
+
cv_skills_soft=cv_skills_soft or [],
|
| 229 |
+
cv_category=cv_category or "Other",
|
| 230 |
)
|
| 231 |
return self._predict_single(job_json, cv_data)
|
| 232 |
|
|
|
|
| 359 |
semantic_matches = self._semantic_skill_matches(
|
| 360 |
missing_technical,
|
| 361 |
individual_skills_vecs,
|
| 362 |
+
threshold=0.88
|
| 363 |
)
|
| 364 |
if semantic_matches:
|
| 365 |
matched_technical = sorted(set(matched_technical) | semantic_matches)
|
|
|
|
| 370 |
|
| 371 |
extra = sorted(cv_skills - job_technical - job_soft)[:8]
|
| 372 |
|
| 373 |
+
tech_denom = min(len(job_technical), 12) if job_technical else 1
|
| 374 |
+
soft_denom = min(len(job_soft), 3) if job_soft else 1
|
| 375 |
+
|
| 376 |
tech_score = min(1.0, len(matched_technical) / tech_denom) if job_technical else 0.50
|
| 377 |
soft_score = min(1.0, len(matched_soft) / soft_denom) if job_soft else 0.50
|
| 378 |
|
| 379 |
+
skill = min(1.0, (tech_score * 0.80) + (soft_score * 0.20))
|
| 380 |
|
| 381 |
# ββ Component scores βββββββββββββββββββββββββββββββββββββββββββββ
|
| 382 |
cv_cat_lower = cv_category.strip().lower() if cv_category else ""
|
|
|
|
| 392 |
experience = _experience_fit(cv_years, job.min_years_experience)
|
| 393 |
|
| 394 |
# ββ Weighted aggregation βββββββββββββββββββββββββββββββββββββββββ
|
|
|
|
|
|
|
|
|
|
| 395 |
base_final = (
|
| 396 |
0.42 * skill
|
| 397 |
+
+ 0.23 * semantic
|
| 398 |
+ 0.12 * title
|
| 399 |
+ 0.11 * cat_score
|
| 400 |
+ 0.07 * seniority
|
|
|
|
| 530 |
cv_vec = self._embed_title(cv_title)
|
| 531 |
job_vec = self._embed_title(job_title)
|
| 532 |
return max(0.2, min(1.0, _cosine_similarity(cv_vec, job_vec)))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|