mahrooshishaq commited on
Commit
7c66219
·
1 Parent(s): 97e667f

Improve sparse PDF extraction

Browse files
Files changed (1) hide show
  1. services/pdf_extractor.py +32 -0
services/pdf_extractor.py CHANGED
@@ -50,6 +50,8 @@ def extract_resume_from_stream(stream) -> dict:
50
  font_sizes.append(float(size))
51
 
52
  page_text = page.extract_text() or ""
 
 
53
  text_pages.append(page_text)
54
 
55
  full_text = "\n".join(text_pages).strip()
@@ -185,6 +187,36 @@ def _detect_multiple_columns(page) -> bool:
185
  return split_rows >= max(4, round(len(rows) * 0.15))
186
 
187
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
188
  def _detect_sections(text: str) -> dict:
189
  """
190
  Splits the resume text into named sections by detecting section headers.
 
50
  font_sizes.append(float(size))
51
 
52
  page_text = page.extract_text() or ""
53
+ if _is_sparse_page_text(page_text, page):
54
+ page_text = _extract_words_as_lines(page) or page_text
55
  text_pages.append(page_text)
56
 
57
  full_text = "\n".join(text_pages).strip()
 
187
  return split_rows >= max(4, round(len(rows) * 0.15))
188
 
189
 
190
+ def _is_sparse_page_text(page_text: str, page) -> bool:
191
+ """True when the regular extractor produced much less text than the word layer."""
192
+ words = page.extract_words(use_text_flow=False, keep_blank_chars=False)
193
+ if not words:
194
+ return False
195
+ word_chars = sum(len(str(word.get("text", ""))) for word in words)
196
+ return len(page_text.strip()) < max(80, word_chars * 0.35)
197
+
198
+
199
+ def _extract_words_as_lines(page) -> str:
200
+ """Rebuild readable lines from positioned words when extract_text is sparse."""
201
+ words = page.extract_words(use_text_flow=False, keep_blank_chars=False)
202
+ if not words:
203
+ return ""
204
+
205
+ rows: dict[int, list[dict]] = {}
206
+ for word in words:
207
+ row_key = round(float(word.get("top", 0)) / 4)
208
+ rows.setdefault(row_key, []).append(word)
209
+
210
+ lines = []
211
+ for _, row_words in sorted(rows.items(), key=lambda item: item[0]):
212
+ ordered = sorted(row_words, key=lambda word: float(word.get("x0", 0)))
213
+ line = " ".join(str(word.get("text", "")).strip() for word in ordered).strip()
214
+ if line:
215
+ lines.append(line)
216
+
217
+ return "\n".join(lines)
218
+
219
+
220
  def _detect_sections(text: str) -> dict:
221
  """
222
  Splits the resume text into named sections by detecting section headers.