File size: 11,321 Bytes
35646e4
275cb5c
85242e3
e11a9ad
275cb5c
85242e3
df1d611
85242e3
df1d611
275cb5c
6b0c8b8
df1d611
e11a9ad
275cb5c
 
6b0c8b8
 
85242e3
6b0c8b8
85242e3
e11a9ad
6b0c8b8
85242e3
b61a150
 
 
85242e3
df1d611
32f64de
 
 
df1d611
b61a150
 
 
 
 
 
 
 
 
85242e3
6b0c8b8
85242e3
6b0c8b8
 
 
85242e3
 
df1d611
e11a9ad
 
 
df1d611
e11a9ad
275cb5c
 
85242e3
df1d611
85242e3
 
df1d611
85242e3
 
df1d611
b61a150
 
 
85242e3
 
 
 
df1d611
85242e3
 
df1d611
85242e3
 
df1d611
85242e3
 
 
 
df1d611
85242e3
 
 
b61a150
85242e3
 
6d87461
e11a9ad
6d87461
df1d611
275cb5c
499bde3
 
 
df1d611
35646e4
df1d611
 
 
499bde3
 
df1d611
6d87461
e11a9ad
499bde3
 
6d87461
 
 
df1d611
e11a9ad
499bde3
 
 
 
 
 
 
85242e3
499bde3
 
 
 
 
 
 
 
 
 
 
df1d611
 
 
499bde3
df1d611
 
499bde3
 
 
 
 
 
 
 
df1d611
499bde3
df1d611
6d87461
e11a9ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
df1d611
 
 
 
 
 
 
 
b61a150
 
 
 
 
 
 
 
 
 
 
 
 
f2fb7ac
b61a150
 
 
f2fb7ac
df1d611
35646e4
dd8eaa7
f2fb7ac
 
 
 
 
 
 
 
 
 
 
dd8eaa7
f2fb7ac
 
 
 
 
 
df1d611
f2fb7ac
 
 
 
 
 
 
85242e3
f2fb7ac
 
85242e3
35646e4
85242e3
df1d611
85242e3
 
b61a150
85242e3
 
 
 
 
 
 
 
 
 
 
35646e4
 
 
85242e3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
df1d611
85242e3
35646e4
85242e3
e11a9ad
df1d611
b61a150
df1d611
85242e3
 
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
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
import re
import fitz  # PyMuPDF
import unicodedata
from gen_ai_hub.proxy.langchain.openai import ChatOpenAI

# ==========================================================
# 1️⃣ TEXT EXTRACTION (Clean + TOC Detection)
# ==========================================================
def extract_text_from_pdf(file_path: str):
    """
    Extracts and cleans text from a PDF using PyMuPDF.
    Handles layout artifacts, numbered sections, and TOC.
    Returns clean text + TOC list + source label.
    """
    text = ""
    try:
        with fitz.open(file_path) as pdf:
            for page_num, page in enumerate(pdf, start=1):
                page_text = page.get_text("text").strip()

                # Fallback: for scanned or weird layouts
                if not page_text:
                    blocks = page.get_text("blocks")
                    page_text = " ".join(
                        block[4] for block in blocks if isinstance(block[4], str)
                    )

                # Ensure bullets & numbered sections start on new lines
                page_text = page_text.replace("• ", "\n• ")
                page_text = re.sub(r"(\d+\.\d+\.\d+)", r"\n\1", page_text)

                # Remove headers/footers and confidential tags
                page_text = re.sub(
                    r"Page\s*\d+\s*(of\s*\d+)?", "", page_text, flags=re.IGNORECASE
                )
                page_text = re.sub(
                    r"(PUBLIC|Confidential|© SAP.*|\bSAP\b\s*\d{4})",
                    "",
                    page_text,
                    flags=re.IGNORECASE,
                )

                text += page_text + "\n"

    except Exception as e:
        raise RuntimeError(f"❌ PDF extraction failed: {e}")

    # --- Cleaning pipeline ---
    text = clean_text(text)

    # --- TOC extraction (Hybrid) ---
    toc, toc_source = get_hybrid_toc(text)
    print(f"📘 TOC Source: {toc_source} | Entries: {len(toc)}")

    return text, toc, toc_source


# ==========================================================
# 2️⃣ ADVANCED CLEANING PIPELINE
# ==========================================================
def clean_text(text: str) -> str:
    """Cleans noisy PDF text before chunking and embedding."""
    text = unicodedata.normalize("NFKD", text)

    # Remove TOC noise (like "6.3.1 Prerequisites .............. 53")
    text = re.sub(
        r"\b\d+(\.\d+){1,}\s+[A-Za-z].{0,40}\.{2,}\s*\d+\b", "", text
    )

    # Replace bullet symbols and dots with consistent spacing
    text = text.replace("•", "- ").replace("▪", "- ").replace("‣", "- ")

    # Remove excessive dots, hyphens, headers
    text = re.sub(r"\.{3,}", ". ", text)
    text = re.sub(r"-\s*\n", "", text)
    text = re.sub(r"\n\s*(PUBLIC|PRIVATE|Confidential)\s*\n", "\n", text, flags=re.IGNORECASE)
    text = re.sub(r"©\s*[A-Z].*?\d{4}", "", text)

    # Normalize newlines and spaces
    text = text.replace("\r", " ")
    text = re.sub(r"\n{2,}", "\n", text)
    text = re.sub(r"\s{2,}", " ", text)

    # Clean leftover special chars
    text = re.sub(r"[^A-Za-z0-9,;:.\-\(\)/&\n\s]", "", text)
    text = re.sub(r"(\s*\.\s*){3,}", " ", text)

    return text.strip()


# ==========================================================
# 3️⃣ TABLE OF CONTENTS DETECTION (Heuristic)
# ==========================================================
def extract_table_of_contents(text: str):
    """
    Smart TOC detector for enterprise PDFs.
    Handles 'Table of Contents', 'Contents', 'Content', 'Index', 'Overview',
    and implicit numbered TOCs without a header.
    Returns list of (section_number, section_title).
    """
    toc_entries = []
    lines = text.split("\n")
    toc_started = False
    toc_ended = False
    line_count = len(lines)

    for i, line in enumerate(lines):
        # --- Step 1️⃣: Detect TOC header variants ---
        if not toc_started and re.search(r"\b(table\s*of\s*contents?|contents?|index|overview)\b", line, re.IGNORECASE):
            next_lines = lines[i + 1 : i + 8]
            if any(re.match(r"^\s*\d+(\.\d+)*\s+[A-Za-z]", l) for l in next_lines):
                toc_started = True
                continue

        # --- Step 2️⃣: Smart fallback — detect implicit TOC ---
        if not toc_started and re.match(r"^\s*\d+(\.\d+)*\s+[A-Za-z]", line):
            numbered_lines = 0
            for j in range(i, min(i + 5, line_count)):
                if re.match(r"^\s*\d+(\.\d+)*\s+[A-Za-z]", lines[j]):
                    numbered_lines += 1
            if numbered_lines >= 3:  # heuristic to confirm pattern
                toc_started = True

        # --- Step 3️⃣: Detect end of TOC region ---
        if toc_started and re.match(r"^\s*(Step\s*\d+|[A-Z][a-z]{2,}\s[A-Z])", line):
            toc_ended = True
            break

        # --- Step 4️⃣: Extract TOC entries ---
        if toc_started and not toc_ended:
            match = re.match(
                r"^\s*(\d+(?:\.\d+)*)\s+([A-Z][A-Za-z0-9\s/&(),-]+)(?:\.+\s*\d+)?$",
                line.strip()
            )
            if match:
                section = match.group(1).strip()
                title = match.group(2).strip()
                if len(title) > 3 and not re.match(r"^\d+$", title):
                    toc_entries.append((section, title))

    # --- Step 5️⃣: Clean up duplicates ---
    deduped = []
    seen = set()
    for sec, title in toc_entries:
        key = (sec, title.lower())
        if key not in seen:
            deduped.append((sec, title))
            seen.add(key)

    return deduped


# ==========================================================
# 3A️⃣ HYBRID TOC FALLBACK (AI-Inferred)
# ==========================================================
def adaptive_fallback_toc(text: str, model: str = "gpt-4o-mini", max_chars: int = 7000):
    """
    Uses an LLM to infer a Table of Contents from the document text.
    Called only when no TOC is found via regex parsing.
    """
    snippet = text[:max_chars]
    llm = ChatOpenAI(model=model, temperature=0)
    prompt = f"""
    You are a document structure analyzer.
    Read the following text and infer its main section titles.
    Output a clean, numbered list (1., 2., 3.) with 5–10 entries max.

    TEXT SAMPLE:
    {snippet}
    """
    try:
        response = llm.invoke(prompt)
        lines = [
            re.sub(r"^[0-9.\-•\\s]+", "", l.strip())
            for l in response.content.splitlines()
            if l.strip()
        ]
        toc_ai = [(str(i + 1), l) for i, l in enumerate(lines) if len(l) > 3]
        return toc_ai
    except Exception as e:
        print(f"⚠️ AI TOC fallback failed: {e}")
        return []


# ==========================================================
# 3B️⃣ UNIFIED WRAPPER (Heuristic + AI Hybrid)
# ==========================================================
def get_hybrid_toc(text: str):
    """
    Attempts heuristic TOC extraction; if none found,
    triggers adaptive AI fallback.
    Returns (toc_entries, source_label).
    """
    toc_entries = extract_table_of_contents(text)
    if toc_entries:
        print(f"📘 TOC detected with {len(toc_entries)} entries (heuristic).")
        return toc_entries, "heuristic"

    print("⚠️ No TOC detected — invoking adaptive AI fallback...")
    toc_ai = adaptive_fallback_toc(text)
    if toc_ai:
        print(f"✨ AI-inferred TOC generated with {len(toc_ai)} entries.")
        return toc_ai, "ai_inferred"

    print("❌ No TOC could be detected or inferred.")
    return [], "none"


# ==========================================================
# 4️⃣ SMART CHUNKING (Auto-Sized + Continuity-Aware)
# ==========================================================
def chunk_text(text: str, chunk_size: int = None, overlap: int = None) -> list:
    """
    Enhanced chunking for structured enterprise PDFs.
    Auto-selects chunk size and keeps procedural context intact.
    """
    text_length = len(text)
    if chunk_size is None:
        if text_length > 200000:
            chunk_size, overlap = 2000, 250
        elif text_length > 50000:
            chunk_size, overlap = 1500, 200
        else:
            chunk_size, overlap = 1000, 150
    elif overlap is None:
        overlap = 150

    print(f"⚙️ Auto-selected chunk_size={chunk_size}, overlap={overlap} (len={text_length})")

    text = re.sub(r"\s+", " ", text.strip())
    section_pattern = (
        r"(?=(?:\n?\d+(?:\.\d+){0,3}\s+[A-Z][^\n]{3,100})|(?:Step\s*\d+[:.\s]))"
    )
    sections = re.split(section_pattern, text)
    sections = [s.strip() for s in sections if s.strip()]

    chunks = []
    for section in sections:
        section = re.sub(r"\n\s*[-•▪‣]\s*", " • ", section)
        bullets = re.split(r"(?=\s*[-•▪‣]\s)", section)
        bullets = [b.strip() for b in bullets if b.strip()]

        if len(bullets) > 2:
            combined = " ".join(bullets)
            if len(combined) > chunk_size * 1.5:
                for i in range(0, len(bullets), 6):
                    block = " ".join(bullets[i:i+6])
                    chunks.append(block.strip())
            else:
                chunks.append(combined.strip())
        else:
            chunks.extend(_split_by_sentence(section, chunk_size, overlap))

    chunks = _merge_small_chunks(chunks, min_len=200)

    # Add continuity overlap
    final_chunks = []
    for i, ch in enumerate(chunks):
        if i == 0:
            final_chunks.append(ch)
        else:
            prev_tail = chunks[i - 1][-overlap:] if overlap > 0 else ""
            final_chunks.append((prev_tail + " " + ch).strip())

    print(f"✅ Final chunks created (continuity-aware): {len(final_chunks)}")
    return final_chunks


# ==========================================================
# 5️⃣ Helper Functions
# ==========================================================
def _split_by_sentence(text, chunk_size=800, overlap=80):
    sentences = re.split(r"(?<=[.!?])\s+", text)
    chunks, current = [], ""
    for sent in sentences:
        if len(current) + len(sent) + 1 <= chunk_size:
            current += " " + sent
        else:
            if current.strip():
                chunks.append(current.strip())
            overlap_part = current[-overlap:] if overlap > 0 else ""
            current = overlap_part + " " + sent
    if current.strip():
        chunks.append(current.strip())
    return chunks


def _merge_small_chunks(chunks, min_len=150):
    merged, buffer = [], ""
    for ch in chunks:
        if len(ch) < min_len:
            buffer += " " + ch
        else:
            if buffer:
                merged.append(buffer.strip())
                buffer = ""
            merged.append(ch.strip())
    if buffer:
        merged.append(buffer.strip())
    return merged


# ==========================================================
# 6️⃣ DEBUGGING (Manual Run)
# ==========================================================
if __name__ == "__main__":
    pdf_path = "sample.pdf"
    text, toc, source = extract_text_from_pdf(pdf_path)
    print("\n📚 TOC Preview:", toc[:5])
    chunks = chunk_text(text)
    print(f"\n✅ {len(chunks)} chunks created.")
    for i, c in enumerate(chunks[:5], 1):
        print(f"\n--- Chunk {i} ---\n{c[:500]}...\n")