Spaces:
Running
Running
File size: 9,408 Bytes
32e4c72 b6a6194 9b4cf19 32e4c72 9b4cf19 32e4c72 b6a6194 32e4c72 9b4cf19 32e4c72 bdc7d9a c379b57 32e4c72 b6a6194 32e4c72 b6a6194 32e4c72 b6a6194 5eb3a63 09a324c 9b4cf19 09a324c 32e4c72 b6a6194 9b4cf19 09a324c 9b4cf19 5820bde 09a324c 5820bde 9b4cf19 09a324c 9b4cf19 09a324c 9b4cf19 32e4c72 9b4cf19 09a324c 9b4cf19 5820bde 9b4cf19 32e4c72 5eb3a63 32e4c72 9b4cf19 32e4c72 5eb3a63 9b4cf19 32e4c72 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 | import pymupdf
from scripts.models import RegulatoryChange
from scripts.utility_functions import color_mapping, get_best_fuzzy_match
def add_infos_to_pdf_agentic(doc, changes, successful_annotations, extraction_method="Landing AI", nlp_preprocessing=True):
type_counts = {
"addition": 0,
"modification": 0,
"deletion": 0,
"unspecified": 0,
}
for change in changes:
change_type = change.type if hasattr(change, "type") else "unspecified"
if change_type in type_counts:
type_counts[change_type] += 1
else:
type_counts["unspecified"] += 1
summary_text = (
"Regulatory Summary:\n"
f"- Extraction Method: {extraction_method}\n"
f"- Nlp preprocessing: {'yes' if nlp_preprocessing else 'no'}\n"
f"- Total Changes: {len(changes)}, Successful Annotations: {successful_annotations}\n"
f"- Additions: {type_counts.get('addition', 0)}\n"
f"- Deletions: {type_counts.get('deletion', 0)}\n"
f"- Modifications: {type_counts.get('modification', 0)}\n"
)
page = doc.load_page(0)
rect = pymupdf.Rect(10, 10, 550, 150)
page.insert_textbox(
rect,
summary_text,
fontsize=9,
fontname="helv",
align=pymupdf.TEXT_ALIGN_LEFT,
color=(0, 0, 0.7),
overlay=True,
)
metadata = doc.metadata
metadata["title"] = "Annotated " + (
metadata["title"] if metadata["title"] else "PDF"
)
metadata["author"] = "Fortiss ReguLens" + (
" & " + metadata["author"] if metadata["author"] else ""
)
metadata["subject"] = "Annotated PDF with regulatory changes"
metadata["keywords"] = "regulatory, changes, annotations, pdf"
doc.set_metadata(metadata)
def add_failed_annotations_to_pdf_agentic(doc, failed_annotations):
"""
Doc is edited in place.
Adds failed annotations to the end of the PDF document.
:param doc: The PyMuPDF document object.
:type doc: pymupdf.Document
:param failed_annotations: The failed annotations to be added.
:type failed_annotations: array
"""
if not failed_annotations:
return
page = doc.new_page(pno=-1)
annotation_str = "Failed Annotations:\n"
for failed_annotation in failed_annotations:
text = failed_annotation["change"].text
change_type = failed_annotation["change"].type
change_str = failed_annotation["change"].category
page_num = failed_annotation["page"]
annotation_str += (
f"Page {page_num}: {text} ({change_type}) Change: {change_str}\n"
)
rect = pymupdf.Rect(20, 20, 580, 822)
page.insert_textbox(
rect,
annotation_str,
fontsize=9,
fontname="helv",
align=pymupdf.TEXT_ALIGN_LEFT,
color=(0, 0, 0.7),
)
def agentic_pdf_annotator(changes: list[RegulatoryChange], file_bytes, extraction_method="Landing AI", nlp_preprocessing=True):
changes = [
c for c in changes if c.confirmed and c.validated
]
if not changes:
return ""
successful_annotations = 0
failed_annotations = []
try:
doc = pymupdf.open(stream=file_bytes, filetype="pdf")
except Exception as e:
return ""
# Sort by length of relevant_text in descending order to avoid overlapping highlights
changes = sorted(changes, key=lambda c: -len(c.text))
annotated_areas = {}
# OPTIMIZATION: Pre-cache all pages and their text content
page_cache = {}
page_text_cache = {}
full_text = ""
for page_num in range(len(doc)):
page = doc[page_num]
page_cache[page_num] = page
page_text = page.get_text()
page_text_cache[page_num] = page_text
full_text += page_text
for change in changes:
page_num = int(change.grounding[0].page)
text = change.text
change_type = change.type
change_str = change.category
comment = change.context
if page_num < 0 or page_num >= len(doc):
results = []
for pnr in range(len(doc)): # search all pages
annotated_areas.setdefault(f"{pnr}", [])
page = page_cache[pnr] # Use cached page
text_instances = page.search_for(text)
for inst in text_instances:
page_num = pnr# remove?
results.append({"page": pnr, "bbox": inst})
results = list(
filter(
lambda result: not any(
result["bbox"].intersects(area)
for area in annotated_areas[f"{result['page']}"]
),
results,
)
)
if not results:
best_match = get_best_fuzzy_match(full_text, change)
if best_match and len(best_match) > 0:
print("found best fuzzy match: ", best_match)
for page_num in range(len(doc)): # search all pages
page = page_cache[page_num] # Use cached page
text_instances = page.search_for(best_match)
for inst in text_instances:
results.append({"page": page_num, "bbox": inst})
# we only want the results that do not overlap with already annotated areas
results = list(
filter(
lambda result: not any(
result["bbox"].intersects(area)
for area in annotated_areas[f"{result['page']}"]
),
results,
)
)
if results: # "flattenning" the results
page_num = results[0]["page"]
doc_page = page_cache[page_num] # Use cached page
results = [r["bbox"] for r in results if r["page"] == page_num]
else:
doc_page = page_cache[page_num] # Use cached page
annotated_areas.setdefault(f"{page_num}", [])
# Search for the relevant text on the page
results = doc_page.search_for(text)
# we only want the results that do not overlap with already annotated areas
results = list(
filter(
lambda result: not any(
result.intersects(area)
for area in annotated_areas[f"{page_num}"]
),
results,
)
)
if not results:
best_match = get_best_fuzzy_match(
page_text_cache[page_num], change # Use cached text
)
if best_match and len(best_match) > 0:
results = doc_page.search_for(best_match)
print("found best fuzzy match: ", best_match)
# we only want the results that do not overlap with already annotated areas
results = list(
filter(
lambda result: not any(
result.intersects(area)
for area in annotated_areas[f"{page_num}"]
),
results,
)
)
if not results:
print(f"No non-overlapping match found on page {page_num} for: '{text}'")
failed_annotations.append({"change": change, "page": page_num})
continue
color = color_mapping.get(change_type, (1, 1, 0))
annotated_areas[f"{page_num}"].append(results[0])
highlight = doc_page.add_highlight_annot(results[0])
highlight.set_colors({"stroke": color})
highlight.set_info(
info={
"title": "Comment",
"content": f"{change_type} - {change_str}\n{comment}",
"name": change_type,
}
)
highlight.update()
successful_annotations += 1
# if the resulting rects contain anything other than our search text we know it is a multiline highlight because for each line
# we will have a new result rect. We need to check if the text in the rect is not equal to our search text but is inside of it
for result in results[1:]:
resulttext = doc_page.get_textbox(result)
if (
(resulttext.strip() != text.strip())
& (resulttext.strip() in text.strip())
# & (
# not any(
# result.intersects(area)
# for area in annotated_areas[f"{page_num}"]
# )
# )
):
highlight = doc_page.add_highlight_annot(result)
highlight.set_colors({"stroke": color})
highlight.update()
annotated_areas[f"{page_num}"].append(result)
add_infos_to_pdf_agentic(doc, changes, successful_annotations, extraction_method, nlp_preprocessing)
add_failed_annotations_to_pdf_agentic(doc, failed_annotations)
result_bytes = doc.tobytes()
return result_bytes
|