ai-talent-finder-backend / ai_module /nlp /resume_ner_extractor.py
ilyass yani
Deploiement backend dans HF Spaces
9df97a2
Raw
History Blame
54 kB
"""
Resume NER Extractor - Enhanced Edition
Robust pattern matching for real-world CV extraction
"""
import re
import logging
import unicodedata
from typing import Dict, List, Set, Any
logger = logging.getLogger(__name__)
class ResumeNERExtractor:
"""
Extract resume entities using advanced pattern matching
Designed for real CVs with complex formatting
"""
# Comprehensive skill database (100+ tech skills)
TECH_SKILLS = {
# Languages
"python", "java", "javascript", "typescript", "csharp", "c++", "php",
"ruby", "go", "rust", "kotlin", "swift", "scala", "r", "matlab",
"perl", "groovy", "c", "cpp", "objective-c", "lua", "shell",
# Web Frontend
"react", "vue", "angular", "nextjs", "gatsby", "svelte", "ember",
"backbone", "html", "css", "sass", "less", "bootstrap", "tailwind",
"webpack", "vite", "rollup", "gulp", "grunt",
# Backend & Frameworks
"fastapi", "django", "flask", "spring", "springboot", "rails", "laravel",
"nodejs", "express", "koa", "hapi", "nestjs", "strapi", "graphql",
# Databases
"sql", "postgresql", "mysql", "mongodb", "redis", "cassandra", "dynamodb",
"firebase", "elasticsearch", "oracle", "mariadb", "sqlite", "neo4j",
# Cloud & DevOps
"aws", "azure", "gcp", "docker", "kubernetes", "openshift", "terraform",
"ansible", "jenkins", "gitlab", "github", "circleci", "travisci",
"heroku", "vercel", "netlify", "cloudinary", "s3", "ec2", "rds",
# Data & AI
"machine learning", "deep learning", "nlp", "tensorflow", "pytorch",
"keras", "sklearn", "pandas", "numpy", "scipy", "matplotlib", "seaborn",
"jupyter", "spark", "hadoop", "hive", "pig", "airflow", "etl",
"data science", "data analysis", "data engineering", "statistics",
# Mobile
"ios", "android", "swift", "kotlin", "react native", "flutter", "xamarin",
# API & Protocols
"rest", "restful", "graphql", "grpc", "soap", "websocket", "mqtt",
"api", "http", "https", "json", "xml", "yaml",
# Version Control
"git", "github", "gitlab", "bitbucket", "svn", "mercurial",
# Other Tools
"jira", "confluence", "slack", "discord", "figma", "sketch",
"vim", "vscode", "intellij",
}
# Job title keywords
JOB_KEYWORDS = {
"senior", "lead", "principal", "architect", "manager", "director",
"head of", "vp", "chief", "officer", "developer", "engineer",
"programmer", "coder", "specialist", "analyst", "consultant",
"full stack", "fullstack", "frontend", "backend", "mobile",
"data scientist", "ml engineer", "devops", "commercial", "assistant",
"responsable", "conseiller", "charge", "chargé", "ingenieur", "ingénieur",
"developpeur", "développeur", "vendeur", "gestionnaire", "stagiaire",
}
# Company indicators
COMPANY_SUFFIXES = {
"inc", "incorporated", "ltd", "limited", "llc", "corp", "corporation",
"company", "co", "gmbh", "ag", "sa", "bv", "pty", "pte",
}
# Education keywords
EDUCATION_KEYWORDS = {
"bachelor", "master", "phd", "doctorate", "computer science",
"engineering", "degree", "diploma", "certificate",
"university", "college", "school", "institute", "academy",
"mba", "postgraduate", "bsc", "msc", "btech", "mtech",
"licence", "bts", "dut", "universite", "université",
"ecole", "école", "esup", "sorbonne",
}
EXPERIENCE_HEADERS = {
"experience", "experiences", "experience professionnelle",
"experiences professionnelles", "professional experience", "work experience",
"expérience", "expériences", "stage", "stages",
}
EDUCATION_HEADERS = {
"education", "formation", "formations", "etudes", "études",
"academic background", "education and training",
}
STOP_HEADERS = {
"skills", "competences", "compétences", "technical skills", "langues",
"languages", "profil", "profile", "contact", "summary", "resume", "résumé",
"centres d'interet", "centres d’intérêt", "interets", "intérêts",
"projects", "projets", "certifications", "hobbies",
}
SOFT_SKILLS_HEADERS = {
"competences", "compétences", "soft skills", "skills",
}
INTEREST_HEADERS = {
"centres d interet", "centres d intérêt", "centres d'interet",
"interets", "intérêts", "hobbies",
}
PROJECT_HEADERS = {
"projects", "project", "projets", "projet", "realisations", "réalisations",
}
CERTIFICATION_HEADERS = {
"certifications", "certification", "certificats", "certificat", "licenses", "licences",
}
PROFILE_HEADERS = {
"profil", "profile", "summary", "professional summary",
}
LANGUAGE_NAMES = {
"francais", "anglais", "espagnol", "allemand", "italien", "portugais",
"arabe", "chinois", "japonais", "russe", "portuguese", "english",
"french", "spanish", "german", "italian",
}
EXPERIENCE_DATE_PATTERN = re.compile(
r"(?:(?:19|20)\d{2})\s*(?:-|–|—|/|to|a|à)\s*(?:(?:19|20)\d{2}|present|current|aujourd hui|aujourdhui|now)|(?:(?:19|20)\d{2})",
flags=re.IGNORECASE,
)
def __init__(self):
"""Initialize extractor"""
self.available = True
self.model_name = "fallback-enhanced-regex"
logger.info("✅ Enhanced NER Extractor initialized")
def extract(self, text: str) -> Dict[str, List[str]]:
"""Extract all entities from text"""
if not text or len(text.strip()) < 20:
return self._empty_entities()
grouped = self._empty_entities()
text_lower = text.lower()
# Extract all entity types
grouped["email"] = self._extract_emails(text)
grouped["phone"] = self._extract_phones(text)
grouped["linkedin"] = self._extract_linkedin_urls(text)
grouped["github"] = self._extract_github_urls(text)
grouped["portfolio"] = self._extract_portfolio_urls(text)
grouped["location"] = self._extract_locations(text)
grouped["name"] = self._extract_names(text)
grouped["job_title"] = self._extract_job_titles(text, text_lower)
grouped["company"] = self._extract_companies(text)
grouped["education"] = self._extract_education(text, text_lower)
grouped["languages"] = self._extract_languages(text)
grouped["soft_skills"] = self._extract_soft_skills(text)
grouped["interests"] = self._extract_interests(text)
grouped["certifications"] = self._extract_certifications(text)
grouped["projects"] = self._extract_projects(text)
grouped["profile_summary"] = self._extract_profile_summary(text)
grouped["skills"] = self._extract_skills(text_lower)
# Deduplicate and clean
for key in grouped:
grouped[key] = list(set(
s.strip() for s in grouped[key]
if s and len(s.strip()) > 0
))
return grouped
def _extract_emails(self, text: str) -> List[str]:
"""Extract email addresses"""
emails = re.findall(
r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b',
text
)
return emails[:5]
def _extract_phones(self, text: str) -> List[str]:
"""Extract phone numbers - be strict"""
patterns = [
# International format: +33 6 12 34 56 78 (must be on single line)
r'(\+\d{1,3}\s?[\d\s\-\(\)]{8,})(?:\n|$|[^\d\s\-\(\)])',
# Common French local format: 06 12 34 56 78
r'\b(0\d(?:[\s\.\-]?\d{2}){4})\b',
# Compact local format: 0612345678
r'\b(0\d{9})\b',
# Alternative: +33612345678 (no spaces)
r'(\+\d{10,})',
]
phones = []
for pattern in patterns:
found = re.findall(pattern, text)
for phone in found:
# Extract just the phone part
phone_clean = phone.strip()
# Stop at first newline or special character
phone_clean = re.sub(r'\n.*$', '', phone_clean, flags=re.DOTALL)
phone_clean = re.sub(r'[^\d\s\-\(\)\+].*$', '', phone_clean)
# Filter out date-like patterns (YYYY - YYYY)
if not re.search(r'\d{4}\s*-\s*\d{4}', phone_clean) and len(phone_clean.strip()) >= 10:
phones.append(phone_clean.strip())
if not phones:
# Fallback for OCR text where separators are inconsistent or split across spaces.
compact_text = re.sub(r"[^\d+]+", " ", text)
fallback_patterns = [
r"\+?\d{1,3}\s?\d{1,3}\s?\d{2}\s?\d{2}\s?\d{2}\s?\d{2}",
r"\b0\d(?:\s?\d{2}){4}\b",
r"\b\+?\d{10,14}\b",
]
for pattern in fallback_patterns:
for match in re.findall(pattern, compact_text):
cleaned = re.sub(r"\s+", " ", match).strip()
if len(re.sub(r"\D", "", cleaned)) >= 10:
phones.append(cleaned)
return list(set(phones))[:3] # Only unique phones
def _extract_names(self, text: str) -> List[str]:
"""Extract person names - look at very beginning"""
lines = text.strip().split('\n')
candidates = []
email_fallback = None
emails = self._extract_emails(text)
if emails:
email_fallback = self._infer_name_from_email(emails[0])
for index, raw_line in enumerate(lines[:80]):
line = raw_line.strip().strip('•*-').strip()
normalized = self._normalize_for_matching(line)
if not line or '@' in line or 'http' in normalized or self._is_section_header(normalized):
continue
if any(token in normalized for token in ('linkedin', 'github', 'contact', 'profil', 'profile')):
continue
if any(char.isdigit() for char in line):
continue
words = [word for word in re.split(r'\s+', line) if word]
if not 2 <= len(words) <= 3:
continue
alpha_words = sum(1 for word in words if re.search(r'[A-Za-zÀ-ÿ]', word))
if alpha_words != len(words):
continue
score = 0
if index < 15:
score += 3
if re.fullmatch(r"[A-ZÀ-Ÿ][A-ZÀ-Ÿ'’\-]+(?:\s+[A-ZÀ-Ÿ][A-ZÀ-Ÿ'’\-]+){1,2}", line):
score += 5
elif re.fullmatch(r"[A-ZÀ-Ÿ][A-Za-zÀ-ÿ'’\-]+(?:\s+[A-ZÀ-Ÿ][A-Za-zÀ-ÿ'’\-]+){1,2}", line):
score += 4
else:
score += 2
if any(keyword in normalized for keyword in ('experience', 'formation', 'education', 'profil', 'contact')):
score -= 4
candidates.append((score, line.title()))
# Also evaluate segmented fragments when a line contains separators like pipes or dashes.
for fragment in re.split(r"\s*[|/·•]\s*|\s+-\s+|\s+–\s+|\s+—\s+", line):
fragment = fragment.strip()
fragment_normalized = self._normalize_for_matching(fragment)
if not fragment or fragment == line:
continue
if '@' in fragment or 'http' in fragment_normalized or self._is_section_header(fragment_normalized):
continue
if any(token in fragment_normalized for token in ('linkedin', 'github', 'contact', 'profil', 'profile')):
continue
if any(char.isdigit() for char in fragment):
continue
fragment_words = [word for word in re.split(r'\s+', fragment) if word]
if not 2 <= len(fragment_words) <= 4:
continue
if not all(re.search(r'[A-Za-zÀ-ÿ]', word) for word in fragment_words):
continue
fragment_score = 2
if index < 10:
fragment_score += 2
if re.fullmatch(r"[A-ZÀ-Ÿ][A-ZÀ-Ÿ'’\-]+(?:\s+[A-ZÀ-Ÿ][A-ZÀ-Ÿ'’\-]+){1,3}", fragment):
fragment_score += 4
elif re.fullmatch(r"[A-ZÀ-Ÿ][A-Za-zÀ-ÿ'’\-]+(?:\s+[A-ZÀ-Ÿ][A-Za-zÀ-ÿ'’\-]+){1,3}", fragment):
fragment_score += 3
candidates.append((fragment_score, fragment.title()))
if not candidates:
if email_fallback:
return [email_fallback]
return []
candidates.sort(key=lambda item: (-item[0], len(item[1])))
best_name = candidates[0][1]
if len(best_name) < 3 and email_fallback:
return [email_fallback]
return [best_name]
def _extract_linkedin_urls(self, text: str) -> List[str]:
"""Extract LinkedIn profile URLs."""
urls = re.findall(r'((?:https?://)?(?:www\.)?linkedin\.com/[^\s,;]+)', text, flags=re.IGNORECASE)
cleaned = []
for url in urls:
normalized = url.strip().rstrip(').,;')
if not normalized.lower().startswith("http"):
normalized = f"https://{normalized}"
cleaned.append(normalized)
return list(dict.fromkeys(cleaned))[:3]
def _extract_github_urls(self, text: str) -> List[str]:
"""Extract GitHub profile/repository URLs."""
urls = re.findall(r'((?:https?://)?(?:www\.)?github\.com/[^\s,;]+)', text, flags=re.IGNORECASE)
cleaned = []
for url in urls:
normalized = url.strip().rstrip(').,;')
if not normalized.lower().startswith("http"):
normalized = f"https://{normalized}"
cleaned.append(normalized)
return list(dict.fromkeys(cleaned))[:5]
def _extract_portfolio_urls(self, text: str) -> List[str]:
"""Extract non-social portfolio websites."""
urls = re.findall(r'((?:https?://)?(?:www\.)?[a-z0-9][a-z0-9\-\.]+\.[a-z]{2,}(?:/[^\s,;]*)?)', text, flags=re.IGNORECASE)
blocked_domains = ("linkedin.com", "github.com", "facebook.com", "instagram.com", "twitter.com", "x.com")
cleaned = []
for url in urls:
normalized = url.strip().rstrip(').,;')
lower = normalized.lower()
if any(domain in lower for domain in blocked_domains):
continue
if "@" in normalized:
continue
if not normalized.lower().startswith("http"):
normalized = f"https://{normalized}"
cleaned.append(normalized)
return list(dict.fromkeys(cleaned))[:5]
def _extract_locations(self, text: str) -> List[str]:
"""Extract likely city/country mentions from contact block."""
locations = []
for line in text.split('\n')[:30]:
cleaned = line.strip()
if not cleaned:
continue
if '@' in cleaned or 'linkedin.com' in cleaned.lower() or re.search(r'\d{2}\s?\d{2}', cleaned):
continue
for match in re.findall(r'\b([A-ZÀ-Ÿ][A-Za-zÀ-ÿ\-]+\s*,\s*[A-ZÀ-Ÿ][A-Za-zÀ-ÿ\-]+)\b', cleaned):
locations.append(match.strip())
return list(dict.fromkeys(locations))[:3]
def _extract_job_titles(self, text: str, text_lower: str) -> List[str]:
"""Extract job titles - be specific"""
job_titles = set()
lines = text.split('\n')
in_experience_section = False
for index, raw_line in enumerate(lines):
line = raw_line.strip().strip('•*-').strip()
normalized = self._normalize_for_matching(line)
if not line:
continue
if self._is_experience_header(normalized):
in_experience_section = True
continue
if in_experience_section and self._is_section_header(normalized) and not self._is_experience_header(normalized):
in_experience_section = False
continue
if not in_experience_section:
continue
if self._looks_like_company_line(line):
for next_line in lines[index + 1:index + 4]:
next_clean = next_line.strip().strip('•*-').strip()
next_normalized = self._normalize_for_matching(next_clean)
if not next_clean:
continue
if self._is_section_header(next_normalized) or self._looks_like_company_line(next_clean):
break
if self._looks_like_job_title(next_clean):
job_titles.add(next_clean)
continue
if index + 1 < len(lines):
next_clean = lines[index + 1].strip().strip('•*-').strip()
if next_clean and self._looks_like_company_line(next_clean) and self._looks_like_job_title(line):
job_titles.add(line)
# Fallback: detect title/company pairs globally when OCR breaks section boundaries.
for index, raw_line in enumerate(lines):
line = raw_line.strip().strip('•*-').strip()
if not line:
continue
if self._looks_like_job_title(line) and index + 1 < len(lines):
next_clean = lines[index + 1].strip().strip('•*-').strip()
if next_clean and self._looks_like_company_line(next_clean):
job_titles.add(line)
return list(job_titles)[:5]
def _extract_companies(self, text: str) -> List[str]:
"""Extract company names"""
companies = set()
# Known tech companies
known_companies = {
"google", "microsoft", "apple", "amazon", "facebook", "meta",
"netflix", "spotify", "uber", "airbnb", "tesla", "twitter",
"linkedin", "ibm", "oracle", "salesforce", "adobe", "nvidia",
}
text_lower = text.lower()
in_experience_section = False
for raw_line in text.split('\n'):
line = raw_line.strip()
normalized = self._normalize_for_matching(line)
if not line:
continue
if self._is_experience_header(normalized):
in_experience_section = True
continue
if in_experience_section and self._is_section_header(normalized) and not self._is_experience_header(normalized):
in_experience_section = False
continue
if in_experience_section and self._looks_like_company_line(line):
company = self._extract_company_from_line(line)
if company:
companies.add(company)
# Fallback: extract company lines globally when section headers are merged.
for raw_line in text.split('\n'):
line = raw_line.strip()
if self._looks_like_company_line(line):
company = self._extract_company_from_line(line)
if company:
companies.add(company)
for company in known_companies:
if company in text_lower and company not in {'linkedin', 'github'}:
companies.add(company.title())
return list(companies)[:5]
def _extract_education(self, text: str, text_lower: str) -> List[str]:
"""Extract education"""
education = set()
lines = text.split('\n')
in_education_section = False
for raw_line in lines:
line = raw_line.strip().strip('•*-').strip()
normalized = self._normalize_for_matching(line)
if not line:
continue
if self._is_education_header(normalized):
in_education_section = True
continue
if in_education_section and self._is_stop_header(normalized):
in_education_section = False
continue
if not in_education_section:
continue
if len(line) < 5 or len(line) > 200:
continue
if self._looks_like_education_line(line):
education.add(line)
for keyword in self.EDUCATION_KEYWORDS:
if keyword in text_lower:
for raw_line in lines:
cleaned = raw_line.strip().strip('•*-').strip()
normalized = self._normalize_for_matching(cleaned)
if keyword in normalized and self._looks_like_education_line(cleaned):
education.add(cleaned)
return list(education)[:10]
def _extract_skills(self, text_lower: str) -> List[str]:
"""Extract technical skills"""
skills = set()
tokens = set(re.findall(r"[a-zA-Z0-9\+\#\-\.]+", text_lower))
for skill in self.TECH_SKILLS:
if " " in skill:
if re.search(rf"\b{re.escape(skill)}\b", text_lower):
skills.add(skill.title())
else:
if skill in tokens:
skills.add(skill.title())
if not skills:
lines = text_lower.split('\n')
in_section = False
for raw_line in lines:
line = raw_line.strip().strip('•*- ').strip()
normalized = self._normalize_for_matching(line)
if not line:
continue
if self._line_matches_any_header(normalized, self.SOFT_SKILLS_HEADERS):
in_section = True
continue
if in_section and self._is_section_header(normalized) and not self._line_matches_any_header(normalized, self.SOFT_SKILLS_HEADERS):
in_section = False
continue
if not in_section:
continue
candidates = [candidate.strip() for candidate in re.split(r"[,;/•|]", line) if candidate.strip()]
for candidate in candidates:
candidate_normalized = self._normalize_for_matching(candidate)
if not candidate or len(candidate) < 2 or len(candidate) > 60:
continue
if candidate_normalized in self.LANGUAGE_NAMES:
continue
if '@' in candidate or 'http' in candidate_normalized:
continue
if re.search(r'\b(19|20)\d{2}\b', candidate):
continue
if self._looks_like_company_line(candidate):
continue
skills.add(candidate.title())
if not skills:
soft_skill_vocabulary = {
"sens du contact", "communication", "capacite d adaptation", "capacité d adaptation",
"polyvalence", "logique", "rigueur", "autonomie", "leadership", "organisation",
"gestion de projet", "travail en équipe", "travail en equipe", "esprit d équipe",
"esprit d equipe", "collaboration", "créativité", "creativite", "analyse", "adaptabilité",
}
for candidate in re.split(r"[\n,;/•|]+", text_lower):
cleaned = candidate.strip()
normalized = self._normalize_for_matching(cleaned)
if normalized in soft_skill_vocabulary:
skills.add(cleaned.title())
return list(skills)[:50]
def _extract_languages(self, text: str) -> List[str]:
"""Extract spoken languages, especially from LANGUES/LANGUAGES section."""
languages = set()
lines = text.split('\n')
in_language_section = False
for raw_line in lines:
line = raw_line.strip().strip('•*-').strip()
normalized = self._normalize_for_matching(line)
if not line:
continue
if self._line_matches_any_header(normalized, {"langues", "languages", "langue"}):
in_language_section = True
continue
if in_language_section and self._is_section_header(normalized) and not self._line_matches_any_header(normalized, {"langues", "languages", "langue"}):
in_language_section = False
continue
if in_language_section:
tokens = re.split(r'[,;\s]+', normalized)
for token in tokens:
if token in self.LANGUAGE_NAMES:
languages.add(token.capitalize())
for token in re.findall(r'\b[A-Za-zÀ-ÿ]+\b', self._normalize_for_matching(text)):
if token in self.LANGUAGE_NAMES:
languages.add(token.capitalize())
return list(languages)[:8]
def _extract_soft_skills(self, text: str) -> List[str]:
"""Extract bullet-point soft skills from COMPÉTENCES section."""
skills = []
lines = text.split('\n')
in_section = False
for raw_line in lines:
line = raw_line.strip()
normalized = self._normalize_for_matching(line)
if not line:
continue
if self._line_matches_any_header(normalized, self.SOFT_SKILLS_HEADERS):
in_section = True
continue
if in_section and self._is_section_header(normalized) and not self._line_matches_any_header(normalized, self.SOFT_SKILLS_HEADERS):
in_section = False
continue
if not in_section:
continue
candidate = line.strip('•*- ').strip()
candidate_normalized = self._normalize_for_matching(candidate)
if not candidate or len(candidate) < 2 or len(candidate) > 60:
continue
if candidate_normalized in self.LANGUAGE_NAMES:
continue
if '@' in candidate or 'http' in candidate_normalized:
continue
if candidate.endswith('.'):
continue
skills.append(candidate)
if not skills:
soft_skill_vocabulary = {
"sens du contact", "communication", "capacite d adaptation", "capacité d adaptation",
"polyvalence", "logique", "rigueur", "autonomie", "leadership", "organisation",
}
for raw_line in lines:
candidate = raw_line.strip().strip('•*- ').strip()
normalized = self._normalize_for_matching(candidate)
if normalized in soft_skill_vocabulary:
skills.append(candidate)
return list(dict.fromkeys(skills))[:20]
def _extract_interests(self, text: str) -> List[str]:
"""Extract interests/hobbies from dedicated section."""
interests = []
lines = text.split('\n')
in_section = False
for raw_line in lines:
line = raw_line.strip()
normalized = self._normalize_for_matching(line)
if not line:
continue
if self._line_matches_any_header(normalized, self.INTEREST_HEADERS):
in_section = True
continue
if in_section and self._is_section_header(normalized) and not self._line_matches_any_header(normalized, self.INTEREST_HEADERS):
in_section = False
continue
if not in_section:
continue
candidate = line.strip('•*- ').strip()
if not candidate or len(candidate) < 2 or len(candidate) > 80:
continue
if '@' in candidate or 'http' in candidate.lower():
continue
normalized_candidate = self._normalize_for_matching(candidate)
if re.search(r'\b(19|20)\d{2}\b', candidate):
continue
if any(keyword in normalized_candidate for keyword in self.EDUCATION_KEYWORDS):
continue
if self._looks_like_company_line(candidate):
continue
interests.append(candidate)
return list(dict.fromkeys(interests))[:20]
def _extract_certifications(self, text: str) -> List[str]:
"""Extract certifications from dedicated sections and inline lines."""
certifications = []
lines = text.split('\n')
in_section = False
for raw_line in lines:
line = raw_line.strip()
normalized = self._normalize_for_matching(line)
if not line:
continue
if self._line_matches_any_header(normalized, self.CERTIFICATION_HEADERS):
in_section = True
continue
if in_section and self._is_section_header(normalized):
in_section = False
continue
candidate = line.strip('•*- ').strip()
if len(candidate) < 3 or len(candidate) > 140:
continue
if in_section:
certifications.append(candidate)
continue
if any(token in normalized for token in ("certification", "certificat", "aws certified", "scrum", "pmp", "itil", "toeic", "toefl")):
certifications.append(candidate)
return list(dict.fromkeys(certifications))[:20]
def _extract_projects(self, text: str) -> List[str]:
"""Extract projects from dedicated sections and project-like bullet lines."""
projects = []
lines = text.split('\n')
in_section = False
for raw_line in lines:
line = raw_line.strip()
normalized = self._normalize_for_matching(line)
if not line:
continue
if self._line_matches_any_header(normalized, self.PROJECT_HEADERS):
in_section = True
continue
if in_section and self._is_section_header(normalized):
in_section = False
continue
candidate = line.strip('•*- ').strip()
if len(candidate) < 4 or len(candidate) > 160:
continue
if '@' in candidate or 'http' in normalized:
continue
if in_section:
projects.append(candidate)
continue
if any(token in normalized for token in ("project", "projet", "developed", "implemented", "built", "realise", "réalisé")):
if not self._looks_like_company_line(candidate):
projects.append(candidate)
return list(dict.fromkeys(projects))[:20]
def _extract_profile_summary(self, text: str) -> List[str]:
"""Extract summary paragraph from PROFIL/PROFILE section."""
lines = text.split('\n')
in_section = False
collected = []
for raw_line in lines:
line = raw_line.strip()
normalized = self._normalize_for_matching(line)
if not line:
continue
if self._line_matches_any_header(normalized, self.PROFILE_HEADERS):
in_section = True
continue
if in_section and self._is_section_header(normalized) and normalized not in self.PROFILE_HEADERS:
break
if in_section:
candidate = line.strip('•*- ').strip()
if len(candidate) < 10:
continue
if '@' in candidate or 'linkedin.com' in candidate.lower():
continue
collected.append(candidate)
if len(" ".join(collected)) > 350:
break
if not collected:
# OCR fallback: summary can appear before the PROFIL header.
lines_count = len(lines)
for i, raw_line in enumerate(lines):
seed = raw_line.strip().strip('•*- ').strip()
seed_normalized = self._normalize_for_matching(seed)
if len(seed) < 20:
continue
if any(token in seed_normalized for token in ("contact", "linkedin", "langues", "competences", "experiences", "formation")):
continue
if not any(keyword in seed_normalized for keyword in ("diplome", "diplômé", "experience", "passion", "relationnel")):
continue
chunk = [seed]
for j in range(i + 1, min(i + 5, lines_count)):
follow = lines[j].strip().strip('•*- ').strip()
follow_normalized = self._normalize_for_matching(follow)
if not follow or len(follow) < 8:
continue
if self._is_section_header(follow_normalized):
break
if any(token in follow_normalized for token in ("contact", "linkedin", "langues", "competences", "experiences", "formation")):
break
if '@' in follow or re.search(r'\b0\d(?:[\s\.\-]?\d{2}){4}\b', follow):
break
chunk.append(follow)
joined_chunk = " ".join(chunk).strip()
if len(joined_chunk) >= 40:
return [joined_chunk]
# Fallback: pick a substantial sentence likely from the profile paragraph.
for raw_line in lines:
candidate = raw_line.strip().strip('•*- ').strip()
normalized = self._normalize_for_matching(candidate)
if len(candidate) < 40:
continue
if '@' in candidate or 'linkedin.com' in normalized:
continue
if any(header in normalized for header in ("experience", "formation", "langues", "competences", "centres")):
continue
if any(keyword in normalized for keyword in ("experience", "diplome", "diplômé", "passion", "relationnel")):
return [candidate]
for sentence in re.split(r'[\n\.]+', text):
candidate = sentence.strip().strip('•*- ').strip()
normalized = self._normalize_for_matching(candidate)
if len(candidate) < 40:
continue
if '@' in candidate or 'linkedin.com' in normalized:
continue
if re.search(r'\b0\d(?:[\s\.\-]?\d{2}){4}\b', candidate):
continue
if any(keyword in normalized for keyword in ("je suis", "experience", "diplome", "diplômé", "relationnel")):
return [candidate]
return []
joined = " ".join(collected)
if re.search(r'\b0\d(?:[\s\.\-]?\d{2}){4}\b', joined):
filtered = []
for part in collected:
normalized = self._normalize_for_matching(part)
if '@' in part or 'linkedin.com' in normalized:
continue
if re.search(r'\b0\d(?:[\s\.\-]?\d{2}){4}\b', part):
continue
if len(part) < 20:
continue
filtered.append(part)
joined = " ".join(filtered)
if not joined:
lines_count = len(lines)
for i, raw_line in enumerate(lines):
seed = raw_line.strip().strip('•*- ').strip()
seed_normalized = self._normalize_for_matching(seed)
if len(seed) < 20:
continue
if any(token in seed_normalized for token in ("contact", "linkedin", "langues", "competences", "experiences", "formation")):
continue
if not any(keyword in seed_normalized for keyword in ("diplome", "diplômé", "experience", "passion", "relationnel")):
continue
chunk = [seed]
for j in range(i + 1, min(i + 5, lines_count)):
follow = lines[j].strip().strip('•*- ').strip()
follow_normalized = self._normalize_for_matching(follow)
if not follow or len(follow) < 8:
continue
if self._is_section_header(follow_normalized):
break
if any(token in follow_normalized for token in ("contact", "linkedin", "langues", "competences", "experiences", "formation")):
break
if '@' in follow or re.search(r'\b0\d(?:[\s\.\-]?\d{2}){4}\b', follow):
break
chunk.append(follow)
joined_chunk = " ".join(chunk).strip()
if len(joined_chunk) >= 40:
joined = joined_chunk
break
return [joined] if joined else []
def _empty_entities(self) -> Dict[str, List[str]]:
"""Return empty structure"""
return {
"name": [],
"email": [],
"phone": [],
"linkedin": [],
"github": [],
"portfolio": [],
"location": [],
"skills": [],
"languages": [],
"soft_skills": [],
"interests": [],
"certifications": [],
"projects": [],
"profile_summary": [],
"company": [],
"job_title": [],
"education": []
}
def _infer_name_from_email(self, email: str | None) -> str | None:
"""Infer a human-readable name from an email local part."""
if not email or "@" not in email:
return None
local_part = email.split("@", 1)[0].strip()
if not local_part:
return None
pieces = [piece for piece in re.split(r"[._\-+]+", local_part) if piece]
if len(pieces) < 2:
return None
candidate = " ".join(piece.capitalize() for piece in pieces[:3]).strip()
if len(candidate) < 4:
return None
return candidate
def _normalize_for_matching(self, value: str) -> str:
value = unicodedata.normalize("NFKD", value or "")
value = "".join(char for char in value if not unicodedata.combining(char))
value = re.sub(r"[^a-zA-Z0-9]+", " ", value)
return re.sub(r"\s+", " ", value).strip().lower()
def _is_experience_header(self, normalized_line: str) -> bool:
normalized_line = normalized_line.rstrip(':')
return normalized_line in {self._normalize_for_matching(header) for header in self.EXPERIENCE_HEADERS}
def _is_education_header(self, normalized_line: str) -> bool:
normalized_line = normalized_line.rstrip(':')
return normalized_line in {self._normalize_for_matching(header) for header in self.EDUCATION_HEADERS}
def _is_stop_header(self, normalized_line: str) -> bool:
normalized_line = normalized_line.rstrip(':')
return normalized_line in {self._normalize_for_matching(header) for header in self.STOP_HEADERS}
def _is_section_header(self, normalized_line: str) -> bool:
normalized_line = normalized_line.rstrip(':')
return (
self._is_experience_header(normalized_line)
or self._is_education_header(normalized_line)
or self._is_stop_header(normalized_line)
)
def _line_matches_any_header(self, normalized_line: str, headers: Set[str]) -> bool:
"""Return True when a normalized OCR line contains any normalized header phrase."""
normalized_headers = {self._normalize_for_matching(header) for header in headers}
parts = [part.strip() for part in normalized_line.split(' ') if part.strip()]
compact_line = " ".join(parts)
for header in normalized_headers:
if not header:
continue
header_parts = [part for part in header.split(' ') if part]
if len(header_parts) == 1:
if compact_line == header:
return True
continue
if compact_line == header or compact_line.startswith(f"{header} "):
return True
return False
def _looks_like_company_line(self, line: str) -> bool:
normalized = self._normalize_for_matching(line)
if 'linkedin.com' in normalized or 'github.com' in normalized:
return False
if any(keyword in normalized for keyword in ('contact', 'profil', 'profile', 'skills', 'competences', 'compétences')):
return False
return bool(
re.search(r"\b\d{4}\b", line)
or '|' in line
or re.search(r"\s[-–]\s", line)
)
def _extract_company_from_line(self, line: str) -> str:
candidate = line.split('|', 1)[0].strip()
candidate = re.split(r"\s*\(\s*\d{4}.*$", candidate)[0].strip()
candidate = re.split(r"\s*[-–]\s*[A-ZÀ-Ÿ][A-Za-zÀ-ÿ\s&'’.-]{1,40}$", candidate)[0].strip()
candidate = candidate.split(',', 1)[0].strip()
candidate = candidate.strip('•*-').strip()
normalized = self._normalize_for_matching(candidate)
if len(candidate) < 2 or 'linkedin' in normalized or 'github' in normalized:
return ''
if any(keyword in normalized for keyword in ('contact', 'profil', 'profile', 'skills', 'competences', 'compétences')):
return ''
if any(keyword in normalized for keyword in ('universite', 'université', 'university', 'ecole', 'école', 'school', 'college', 'institute', 'esup')):
return ''
return candidate
def _looks_like_job_title(self, line: str) -> bool:
normalized = self._normalize_for_matching(line)
if self._is_section_header(normalized):
return False
if self._looks_like_company_line(line):
return False
if any(token in normalized for token in ('linkedin', 'github', 'http', 'www', 'profil', 'profile')):
return False
if any(keyword in normalized for keyword in self.EDUCATION_KEYWORDS):
return False
if normalized in self.LANGUAGE_NAMES:
return False
if line.endswith('.') or line.endswith(':'):
return False
if any(token in normalized for token in ('bénévolat', 'randonnée', 'voyage', 'théâtre', 'concerts', 'loisirs', 'interets', 'intérêts')):
return False
words = [word for word in re.split(r'\s+', line.strip()) if word]
if not 1 <= len(words) <= 6:
return False
if any(keyword in normalized for keyword in self.JOB_KEYWORDS):
return True
if len(words) == 1:
return len(words[0]) >= 7 and words[0].isalpha()
if all(re.search(r'[A-Za-zÀ-ÿ]', word) for word in words):
return True
return False
def _looks_like_education_line(self, line: str) -> bool:
normalized = self._normalize_for_matching(line)
if self._is_section_header(normalized):
return False
if any(token in normalized for token in ('linkedin', 'github', 'http', 'www')):
return False
if any(token in normalized for token in ('bénévolat', 'randonnée', 'voyage', 'théâtre', 'concerts')):
return False
has_education_keyword = any(keyword in normalized for keyword in self.EDUCATION_KEYWORDS)
if self._looks_like_company_line(line) and not has_education_keyword:
return False
if re.search(r'\b(19|20)\d{2}\b', line):
return True
if has_education_keyword:
return True
return False
def _extract_period(self, line: str) -> str:
"""Extract date range/year from a line."""
match = self.EXPERIENCE_DATE_PATTERN.search(line)
if not match:
return ""
period = re.sub(r"\s+", " ", match.group(0)).strip()
period = period.replace("aujourd hui", "Present").replace("aujourdhui", "Present")
return period
def _strip_period(self, line: str) -> str:
"""Remove date range/year from a line to keep role/company content."""
return self.EXPERIENCE_DATE_PATTERN.sub("", line, count=1).strip(" |-–—,;()")
def _is_likely_experience_anchor(self, line: str) -> bool:
"""Detect lines that likely start a new experience entry."""
if not line or len(line) < 3:
return False
normalized = self._normalize_for_matching(line)
if self._is_section_header(normalized):
return False
if any(token in normalized for token in ("education", "formation", "universite", "university", "ecole", "school")):
return False
has_period = bool(self.EXPERIENCE_DATE_PATTERN.search(line))
has_separator = any(sep in line for sep in ("|", " - ", " – ", " — "))
looks_title = self._looks_like_job_title(line)
looks_company = self._looks_like_company_line(line)
return has_period or (has_separator and (looks_title or looks_company))
def _parse_anchor_line(self, line: str, prev_line: str = "", next_line: str = "") -> Dict[str, Any]:
"""Parse a potential anchor line into title/company/period."""
period = self._extract_period(line)
core = self._strip_period(line)
if not core:
core = line.strip()
segments = [part.strip(" |-–—,;") for part in re.split(r"\||\s+-\s+|\s+–\s+|\s+—\s+", core) if part.strip()]
title = ""
company = ""
for segment in segments:
if not title and self._looks_like_job_title(segment):
title = segment
continue
if not company and self._looks_like_company_line(segment):
parsed = self._extract_company_from_line(segment)
company = parsed or segment
if not company:
company = self._extract_company_from_line(line)
if not title and prev_line and self._looks_like_job_title(prev_line):
title = prev_line
if not title and next_line and self._looks_like_job_title(next_line):
title = next_line
if not company and next_line and self._looks_like_company_line(next_line):
company = self._extract_company_from_line(next_line)
if not company and not title:
return {}
if company and any(keyword in self._normalize_for_matching(company) for keyword in self.EDUCATION_KEYWORDS):
return {}
return {
"title": title,
"company": company,
"period": period or None,
"responsibilities": [],
}
def _extract_experiences(self, text: str) -> List[Dict[str, Any]]:
"""Extract structured professional experiences from varied CV formats."""
original_lines = text.split('\n')
cleaned_lines = [line.strip().strip('•*- ').strip() for line in original_lines]
start_idx = 0
for idx, line in enumerate(cleaned_lines):
normalized = self._normalize_for_matching(line)
if self._line_matches_any_header(normalized, self.EXPERIENCE_HEADERS):
start_idx = idx + 1
break
experiences: List[Dict[str, Any]] = []
current_exp: Dict[str, Any] = {}
def flush_current() -> None:
nonlocal current_exp
if not current_exp:
return
if current_exp.get("company") or current_exp.get("title"):
current_exp["responsibilities"] = list(dict.fromkeys(current_exp.get("responsibilities", [])))[:10]
experiences.append(current_exp)
current_exp = {}
for idx in range(start_idx, len(cleaned_lines)):
line = cleaned_lines[idx]
raw_line = original_lines[idx].strip()
if not line:
continue
normalized = self._normalize_for_matching(line)
if self._is_section_header(normalized) and not self._line_matches_any_header(normalized, self.EXPERIENCE_HEADERS):
if experiences and any(token in normalized for token in ("education", "formation", "skills", "competences", "langues", "languages")):
break
continue
prev_line = cleaned_lines[idx - 1] if idx > 0 else ""
next_line = cleaned_lines[idx + 1] if idx + 1 < len(cleaned_lines) else ""
if self._is_likely_experience_anchor(line):
parsed = self._parse_anchor_line(line, prev_line=prev_line, next_line=next_line)
if parsed:
flush_current()
current_exp = parsed
continue
if not current_exp and self._looks_like_job_title(line):
candidate = {
"title": line,
"company": "",
"period": None,
"responsibilities": [],
}
if next_line and self._looks_like_company_line(next_line):
candidate["company"] = self._extract_company_from_line(next_line)
if candidate["company"]:
current_exp = candidate
continue
if current_exp:
if normalized in self.LANGUAGE_NAMES:
continue
if '@' in line or 'linkedin.com' in normalized:
continue
if self._looks_like_education_line(line):
continue
if not current_exp.get("title") and self._looks_like_job_title(line):
current_exp["title"] = line
continue
if not current_exp.get("company") and self._looks_like_company_line(line):
current_exp["company"] = self._extract_company_from_line(line)
continue
if not current_exp.get("period"):
period = self._extract_period(line)
if period:
current_exp["period"] = period
looks_like_description = (
raw_line.startswith("•")
or raw_line.startswith("-")
or raw_line.startswith("*")
or len(line) > 30
)
if looks_like_description and not self._is_likely_experience_anchor(line):
responsibilities = current_exp.get("responsibilities", [])
if responsibilities and not raw_line.startswith(("•", "-", "*")) and not responsibilities[-1].endswith('.'):
responsibilities[-1] = f"{responsibilities[-1]} {line}".strip()
else:
responsibilities.append(line)
current_exp["responsibilities"] = responsibilities
flush_current()
normalized_seen = set()
deduped_experiences = []
for exp in experiences:
key = (
self._normalize_for_matching(exp.get("title") or ""),
self._normalize_for_matching(exp.get("company") or ""),
(exp.get("period") or "").lower(),
)
if key in normalized_seen:
continue
normalized_seen.add(key)
if not exp.get("company") and not exp.get("title"):
continue
deduped_experiences.append({
"title": exp.get("title", ""),
"company": exp.get("company", ""),
"period": exp.get("period"),
"responsibilities": exp.get("responsibilities", [])[:10],
})
return deduped_experiences[:12]
def extract_structured_profile(self, text: str) -> Dict:
"""Extract and format for API response"""
entities = self.extract(text)
experiences = self._extract_experiences(text)
return {
"full_name": entities["name"][0] if entities["name"] else None,
"email": entities["email"][0] if entities["email"] else None,
"phone": entities["phone"][0] if entities["phone"] else None,
"linkedin_url": entities["linkedin"][0] if entities["linkedin"] else None,
"linkedin_urls": entities["linkedin"][:3],
"github_urls": entities["github"][:5],
"portfolio_urls": entities["portfolio"][:5],
"locations": entities["location"][:5],
"education": entities["education"][:5],
"skills": list(set(entities["skills"])),
"languages": entities["languages"][:8],
"soft_skills": entities["soft_skills"][:20],
"interests": entities["interests"][:20],
"certifications": entities["certifications"][:20],
"projects": entities["projects"][:20],
"profile_summary": entities["profile_summary"][0] if entities["profile_summary"] else None,
"companies": entities["company"][:5],
"job_titles": entities["job_title"][:5],
"experiences": experiences,
"extraction_metadata": {
"model": self.model_name,
"model_available": self.available,
"total_entities": sum(len(v) for v in entities.values())
}
}