LLM-ChatBot-Document / utils /pdf_processor.py
FauzanAriyatmoko's picture
feat: Implement an interactive document viewer with citation highlighting and structured PDF text extraction.
a86d063
"""
PDF Processing utilities for extracting and chunking text from PDF files
"""
import os
from typing import List, Dict
import PyPDF2
import pdfplumber
try:
from langchain_text_splitters import RecursiveCharacterTextSplitter
except ImportError:
from langchain.text_splitter import RecursiveCharacterTextSplitter
from config.model_config import config
class PDFProcessor:
"""Handle PDF text extraction and processing"""
def __init__(self):
self.text_splitter = RecursiveCharacterTextSplitter(
chunk_size=config.CHUNK_SIZE,
chunk_overlap=config.CHUNK_OVERLAP,
length_function=len,
separators=["\n\n", "\n", " ", ""]
)
def extract_text_from_pdf(self, pdf_path: str, method: str = "pdfplumber") -> str:
"""
Extract text from PDF file
Args:
pdf_path: Path to PDF file
method: Extraction method ('pypdf2' or 'pdfplumber')
Returns:
Extracted text as string
"""
text = ""
try:
if method == "pdfplumber":
text = self._extract_with_pdfplumber(pdf_path)
else:
text = self._extract_with_pypdf2(pdf_path)
except Exception as e:
print(f"Error extracting text from {pdf_path}: {e}")
# Fallback to alternative method
if method == "pdfplumber":
text = self._extract_with_pypdf2(pdf_path)
else:
text = self._extract_with_pdfplumber(pdf_path)
return text
def _extract_with_pypdf2(self, pdf_path: str) -> str:
"""Extract text using PyPDF2"""
text = ""
with open(pdf_path, 'rb') as file:
pdf_reader = PyPDF2.PdfReader(file)
for page in pdf_reader.pages:
text += page.extract_text() + "\n"
return text
def _extract_with_pdfplumber(self, pdf_path: str) -> str:
"""Extract text using pdfplumber (better for complex PDFs)"""
text = ""
with pdfplumber.open(pdf_path) as pdf:
for page in pdf.pages:
page_text = page.extract_text()
if page_text:
text += page_text + "\n"
return text
def chunk_text(self, text: str) -> List[str]:
"""
Split text into chunks
Args:
text: Input text to chunk
Returns:
List of text chunks
"""
chunks = self.text_splitter.split_text(text)
return chunks
def extract_with_structure(self, pdf_path: str) -> Dict:
"""
Extract text with page and paragraph structure
Args:
pdf_path: Path to PDF file
Returns:
Dictionary with structured content including pages and paragraphs
"""
structured_content = {
"pages": [],
"paragraphs": [],
"full_text": ""
}
try:
with pdfplumber.open(pdf_path) as pdf:
paragraph_id = 0
for page_num, page in enumerate(pdf.pages, start=1):
page_text = page.extract_text()
if not page_text:
continue
# Split into paragraphs (double newline or significant whitespace)
raw_paragraphs = page_text.split('\n\n')
page_paragraphs = []
for para_text in raw_paragraphs:
para_text = para_text.strip()
if len(para_text) > 20: # Ignore very short fragments
paragraph_id += 1
paragraph_data = {
"id": f"para_{paragraph_id}",
"page": page_num,
"text": para_text,
"char_start": len(structured_content["full_text"]),
"char_end": len(structured_content["full_text"]) + len(para_text)
}
page_paragraphs.append(paragraph_data)
structured_content["paragraphs"].append(paragraph_data)
structured_content["full_text"] += para_text + "\n\n"
structured_content["pages"].append({
"page_num": page_num,
"text": page_text,
"paragraphs": page_paragraphs
})
except Exception as e:
print(f"Error extracting structured content: {e}")
# Fallback to simple extraction
text = self.extract_text_from_pdf(pdf_path)
structured_content["full_text"] = text
structured_content["paragraphs"] = [{
"id": "para_1",
"page": 1,
"text": text,
"char_start": 0,
"char_end": len(text)
}]
return structured_content
def generate_html_preview(self, structured_content: Dict, filename: str) -> str:
"""
Generate HTML representation of PDF for viewer
Args:
structured_content: Structured content from extract_with_structure
filename: Name of the PDF file
Returns:
HTML string
"""
html = f"""
<div class="document-content" data-filename="{filename}">
<div class="document-header">
<h3>📄 {filename}</h3>
<p class="doc-meta">{len(structured_content['pages'])} halaman • {len(structured_content['paragraphs'])} paragraf</p>
</div>
"""
for page in structured_content["pages"]:
html += f"""
<div class="pdf-page" data-page="{page['page_num']}">
<div class="page-number">Halaman {page['page_num']}</div>
"""
for para in page["paragraphs"]:
html += f"""
<p class="paragraph" id="{para['id']}" data-page="{para['page']}">
{para['text']}
</p>
"""
html += "</div>"
html += "</div>"
return html
def chunk_text_with_metadata(self, structured_content: Dict) -> List[Dict]:
"""
Split text into chunks with metadata about source location
Args:
structured_content: Structured content from extract_with_structure
Returns:
List of dictionaries with chunk text and metadata
"""
# Get chunks from the splitter
text_chunks = self.text_splitter.split_text(structured_content["full_text"])
chunks_with_metadata = []
for i, chunk_text in enumerate(text_chunks):
# Find which paragraphs this chunk overlaps with
chunk_start = structured_content["full_text"].find(chunk_text)
chunk_end = chunk_start + len(chunk_text)
# Find overlapping paragraphs
related_paragraphs = []
related_pages = set()
for para in structured_content["paragraphs"]:
# Check if chunk overlaps with paragraph
if not (chunk_end < para["char_start"] or chunk_start > para["char_end"]):
related_paragraphs.append(para["id"])
related_pages.add(para["page"])
chunks_with_metadata.append({
"text": chunk_text,
"chunk_index": i,
"paragraph_ids": related_paragraphs,
"pages": sorted(list(related_pages)),
"char_start": chunk_start,
"char_end": chunk_end
})
return chunks_with_metadata
def process_pdf(self, pdf_path: str) -> Dict:
"""
Complete processing pipeline: extract and chunk PDF with structure
Args:
pdf_path: Path to PDF file
Returns:
Dictionary with filename, text, chunks, and structured content
"""
filename = os.path.basename(pdf_path)
# Extract structured content
structured_content = self.extract_with_structure(pdf_path)
if not structured_content["full_text"].strip():
raise ValueError(f"No text extracted from {filename}")
# Generate HTML preview
html_preview = self.generate_html_preview(structured_content, filename)
# Chunk text with metadata
chunks_with_metadata = self.chunk_text_with_metadata(structured_content)
# Extract just the text for backward compatibility
chunks = [c["text"] for c in chunks_with_metadata]
return {
"filename": filename,
"full_text": structured_content["full_text"],
"chunks": chunks,
"chunks_metadata": chunks_with_metadata,
"structured_content": structured_content,
"html_preview": html_preview,
"num_chunks": len(chunks),
"total_chars": len(structured_content["full_text"]),
"num_pages": len(structured_content["pages"]),
"num_paragraphs": len(structured_content["paragraphs"])
}
def get_pdf_info(self, pdf_path: str) -> Dict:
"""
Get metadata about PDF file
Args:
pdf_path: Path to PDF file
Returns:
Dictionary with PDF metadata
"""
info = {
"filename": os.path.basename(pdf_path),
"file_size": os.path.getsize(pdf_path),
"num_pages": 0
}
try:
with open(pdf_path, 'rb') as file:
pdf_reader = PyPDF2.PdfReader(file)
info["num_pages"] = len(pdf_reader.pages)
except Exception as e:
print(f"Error getting PDF info: {e}")
return info