advanced-tokenizer-system / intelligent_chunking_processor.py
9x25dillon's picture
Upload folder using huggingface_hub
968c919 verified
#!/usr/bin/env python3
"""
Intelligent Chunking Processor
==============================
Advanced chunking system with semantic awareness and context preservation.
"""
import re
import json
import hashlib
import numpy as np
from typing import List, Dict, Any, Optional, Tuple, Generator
from dataclasses import dataclass, asdict
from datetime import datetime
import spacy
from sentence_transformers import SentenceTransformer
import networkx as nx
from sklearn.cluster import KMeans
from sklearn.metrics.pairwise import cosine_similarity
@dataclass
class ChunkMetadata:
"""Metadata for a text chunk."""
chunk_id: str
content_type: str
semantic_topic: str
importance_score: float
context_connections: List[str]
language: str
readability_score: float
entity_count: int
sentiment_score: float
@dataclass
class IntelligentChunk:
"""Intelligent chunk with semantic metadata."""
chunk_id: str
content: str
chunk_index: int
total_chunks: int
file_hash: str
metadata: ChunkMetadata
semantic_embedding: Optional[np.ndarray] = None
timestamp: str = ""
class IntelligentChunkingProcessor:
"""Advanced chunking processor with semantic awareness."""
def __init__(self,
max_chunk_size: int = 1000000,
overlap_size: int = 1000,
semantic_model: str = "all-MiniLM-L6-v2",
language_model: str = "en_core_web_sm"):
self.max_chunk_size = max_chunk_size
self.overlap_size = overlap_size
# Initialize NLP models
self.semantic_model = None
self.nlp = None
self._load_models(semantic_model, language_model)
# Content type patterns
self.content_patterns = {
'code': [
r'```[\s\S]*?```', # Code blocks
r'`[^`]+`', # Inline code
r'def\s+\w+\s*\(', # Python functions
r'class\s+\w+', # Python classes
r'function\s+\w+\s*\(', # JavaScript functions
r'#include\s*<', # C/C++ includes
],
'mathematical': [
r'\$[\s\S]*?\$', # LaTeX math
r'\\[a-zA-Z]+\{[^}]*\}', # LaTeX commands
r'\b\d+\s*[+\-*/=]\s*\d+', # Simple math
r'\\frac\{[^}]+\}\{[^}]+\}', # Fractions
],
'structured_data': [
r'\{[\s\S]*?\}', # JSON objects
r'\[[\s\S]*?\]', # JSON arrays
r'<[^>]+>', # XML/HTML tags
r'^\s*[a-zA-Z_][a-zA-Z0-9_]*\s*:', # Key-value pairs
],
'natural_language': [
r'[.!?]+\s+[A-Z]', # Sentence boundaries
r'\n\n+', # Paragraph breaks
]
}
def _load_models(self, semantic_model: str, language_model: str):
"""Load NLP models."""
try:
# Load semantic model
self.semantic_model = SentenceTransformer(semantic_model)
print(f"โœ… Loaded semantic model: {semantic_model}")
except Exception as e:
print(f"โš ๏ธ Semantic model loading failed: {e}")
self.semantic_model = None
try:
# Load language model
self.nlp = spacy.load(language_model)
print(f"โœ… Loaded language model: {language_model}")
except Exception as e:
print(f"โš ๏ธ Language model loading failed: {e}")
self.nlp = None
def detect_content_type(self, content: str) -> str:
"""Detect the primary content type of the text."""
content = content.strip()
# Check for code patterns
code_matches = 0
for pattern in self.content_patterns['code']:
code_matches += len(re.findall(pattern, content, re.MULTILINE))
if code_matches > 0:
return 'code'
# Check for mathematical content
math_matches = 0
for pattern in self.content_patterns['mathematical']:
math_matches += len(re.findall(pattern, content))
if math_matches > 0:
return 'mathematical'
# Check for structured data
structured_matches = 0
for pattern in self.content_patterns['structured_data']:
structured_matches += len(re.findall(pattern, content))
if structured_matches > len(content) / 100: # Threshold for structured content
return 'structured_data'
# Default to natural language
return 'natural_language'
def extract_semantic_topics(self, content: str) -> List[str]:
"""Extract semantic topics from content."""
if not self.nlp:
return ['general']
try:
doc = self.nlp(content)
# Extract noun phrases and named entities
topics = []
# Named entities
for ent in doc.ents:
if ent.label_ in ['PERSON', 'ORG', 'GPE', 'EVENT', 'WORK_OF_ART', 'LAW']:
topics.append(ent.text.lower())
# Noun phrases
for chunk in doc.noun_chunks:
if len(chunk.text.split()) >= 2: # Multi-word phrases
topics.append(chunk.text.lower())
# Remove duplicates and limit
topics = list(set(topics))[:10]
return topics if topics else ['general']
except Exception as e:
print(f"โš ๏ธ Topic extraction failed: {e}")
return ['general']
def calculate_importance_score(self, content: str, content_type: str) -> float:
"""Calculate importance score for content."""
score = 0.5 # Base score
# Length factor
length_score = min(len(content) / 1000, 1.0) * 0.2
score += length_score
# Content type factor
type_scores = {
'code': 0.3,
'mathematical': 0.25,
'structured_data': 0.2,
'natural_language': 0.1
}
score += type_scores.get(content_type, 0.1)
# Keyword density
important_keywords = [
'important', 'critical', 'essential', 'key', 'main', 'primary',
'function', 'class', 'method', 'algorithm', 'definition', 'theorem',
'conclusion', 'summary', 'abstract', 'introduction'
]
keyword_count = sum(1 for keyword in important_keywords if keyword.lower() in content.lower())
keyword_score = min(keyword_count / 10, 0.3)
score += keyword_score
return min(score, 1.0)
def calculate_readability_score(self, content: str) -> float:
"""Calculate readability score (simplified Flesch score)."""
if not self.nlp:
return 0.5
try:
doc = self.nlp(content)
sentences = [sent for sent in doc.sents]
words = [token for token in doc if not token.is_punct and not token.is_space]
if not sentences or not words:
return 0.5
avg_sentence_length = len(words) / len(sentences)
avg_syllables_per_word = sum(self._count_syllables(word.text) for word in words) / len(words)
# Simplified Flesch score
score = 206.835 - (1.015 * avg_sentence_length) - (84.6 * avg_syllables_per_word)
# Normalize to 0-1
return max(0, min(1, score / 100))
except Exception as e:
print(f"โš ๏ธ Readability calculation failed: {e}")
return 0.5
def _count_syllables(self, word: str) -> int:
"""Count syllables in a word (simplified)."""
word = word.lower()
vowels = 'aeiouy'
syllable_count = 0
prev_was_vowel = False
for char in word:
if char in vowels:
if not prev_was_vowel:
syllable_count += 1
prev_was_vowel = True
else:
prev_was_vowel = False
# Handle silent 'e'
if word.endswith('e') and syllable_count > 1:
syllable_count -= 1
return max(1, syllable_count)
def calculate_sentiment_score(self, content: str) -> float:
"""Calculate sentiment score (-1 to 1)."""
if not self.nlp:
return 0.0
try:
doc = self.nlp(content)
# Simple sentiment based on positive/negative words
positive_words = ['good', 'great', 'excellent', 'amazing', 'wonderful', 'fantastic', 'perfect']
negative_words = ['bad', 'terrible', 'awful', 'horrible', 'disappointing', 'wrong', 'error']
pos_count = sum(1 for word in doc if word.text.lower() in positive_words)
neg_count = sum(1 for word in doc if word.text.lower() in negative_words)
total_words = len([token for token in doc if token.is_alpha])
if total_words == 0:
return 0.0
sentiment = (pos_count - neg_count) / total_words
return max(-1, min(1, sentiment))
except Exception as e:
print(f"โš ๏ธ Sentiment calculation failed: {e}")
return 0.0
def extract_entities(self, content: str) -> int:
"""Extract and count entities."""
if not self.nlp:
return 0
try:
doc = self.nlp(content)
entities = [ent for ent in doc.ents if ent.label_ in ['PERSON', 'ORG', 'GPE', 'EVENT', 'WORK_OF_ART']]
return len(entities)
except Exception as e:
print(f"โš ๏ธ Entity extraction failed: {e}")
return 0
def detect_language(self, content: str) -> str:
"""Detect language of content."""
if not self.nlp:
return 'en'
try:
doc = self.nlp(content[:1000]) # Sample first 1000 chars
return doc.lang_ if hasattr(doc, 'lang_') else 'en'
except Exception as e:
print(f"โš ๏ธ Language detection failed: {e}")
return 'en'
def generate_semantic_embedding(self, content: str) -> Optional[np.ndarray]:
"""Generate semantic embedding for content."""
if not self.semantic_model:
return None
try:
embedding = self.semantic_model.encode(content)
return embedding
except Exception as e:
print(f"โš ๏ธ Embedding generation failed: {e}")
return None
def find_semantic_boundaries(self, content: str, content_type: str) -> List[int]:
"""Find optimal chunk boundaries based on content type."""
boundaries = []
if content_type == 'code':
# For code, split on function/class boundaries
patterns = [
r'\n\s*(def\s+\w+\s*\(|class\s+\w+|function\s+\w+\s*\()',
r'\n\s*#\s*---+\n', # Comment separators
r'\n\s*//\s*---+\n', # Comment separators
r'\n\n+', # Multiple newlines
]
elif content_type == 'natural_language':
# For natural language, split on paragraph/section boundaries
patterns = [
r'\n\s*#{1,6}\s+', # Markdown headers
r'\n\n+', # Paragraph breaks
r'[.!?]\s+\n', # Sentence ends followed by newline
]
elif content_type == 'structured_data':
# For structured data, split on object/array boundaries
patterns = [
r'\n\s*\{', # New JSON objects
r'\n\s*\[', # New JSON arrays
r'\n\s*<[^>]+>', # New XML/HTML elements
]
else:
# Default patterns
patterns = [r'\n\n+', r'[.!?]\s+\n']
for pattern in patterns:
for match in re.finditer(pattern, content):
boundaries.append(match.start())
# Add beginning and end
boundaries = [0] + sorted(set(boundaries)) + [len(content)]
return boundaries
def create_intelligent_chunks(self,
content: str,
file_hash: str,
chunk_overlap: int = None) -> List[IntelligentChunk]:
"""Create intelligent chunks with semantic awareness."""
if chunk_overlap is None:
chunk_overlap = self.overlap_size
# Detect content type
content_type = self.detect_content_type(content)
# If content is small enough, return as single chunk
if len(content) <= self.max_chunk_size:
metadata = self._create_chunk_metadata(
content, content_type, chunk_index=0, total_chunks=1
)
embedding = self.generate_semantic_embedding(content)
return [IntelligentChunk(
chunk_id="chunk_0",
content=content,
chunk_index=0,
total_chunks=1,
file_hash=file_hash,
metadata=metadata,
semantic_embedding=embedding,
timestamp=datetime.now().isoformat()
)]
# Find semantic boundaries
boundaries = self.find_semantic_boundaries(content, content_type)
# Create chunks based on boundaries and size constraints
chunks = []
total_chunks = 0
# Calculate optimal number of chunks
estimated_chunks = max(1, len(content) // (self.max_chunk_size - chunk_overlap))
total_chunks = estimated_chunks
for i in range(total_chunks):
start_idx = i * (self.max_chunk_size - chunk_overlap)
end_idx = min(start_idx + self.max_chunk_size, len(content))
# Adjust boundaries to semantic boundaries if possible
if boundaries:
# Find the best semantic boundary near our calculated boundary
best_boundary = end_idx
for boundary in boundaries:
if start_idx < boundary < end_idx:
# Prefer boundaries closer to our calculated end
if abs(boundary - end_idx) < abs(best_boundary - end_idx):
best_boundary = boundary
end_idx = best_boundary
chunk_content = content[start_idx:end_idx]
# Create metadata
metadata = self._create_chunk_metadata(
chunk_content, content_type, chunk_index=i, total_chunks=total_chunks
)
# Generate embedding
embedding = self.generate_semantic_embedding(chunk_content)
chunk = IntelligentChunk(
chunk_id=f"chunk_{i}",
content=chunk_content,
chunk_index=i,
total_chunks=total_chunks,
file_hash=file_hash,
metadata=metadata,
semantic_embedding=embedding,
timestamp=datetime.now().isoformat()
)
chunks.append(chunk)
# Update total chunks
for chunk in chunks:
chunk.total_chunks = len(chunks)
return chunks
def _create_chunk_metadata(self, content: str, content_type: str, chunk_index: int, total_chunks: int) -> ChunkMetadata:
"""Create metadata for a chunk."""
# Extract topics
topics = self.extract_semantic_topics(content)
primary_topic = topics[0] if topics else 'general'
# Calculate scores
importance_score = self.calculate_importance_score(content, content_type)
readability_score = self.calculate_readability_score(content)
sentiment_score = self.calculate_sentiment_score(content)
entity_count = self.extract_entities(content)
language = self.detect_language(content)
# Generate context connections (simplified)
context_connections = []
if chunk_index > 0:
context_connections.append(f"chunk_{chunk_index-1}")
if chunk_index < total_chunks - 1:
context_connections.append(f"chunk_{chunk_index+1}")
return ChunkMetadata(
chunk_id=f"chunk_{chunk_index}",
content_type=content_type,
semantic_topic=primary_topic,
importance_score=importance_score,
context_connections=context_connections,
language=language,
readability_score=readability_score,
entity_count=entity_count,
sentiment_score=sentiment_score
)
def cluster_chunks_by_semantics(self, chunks: List[IntelligentChunk], n_clusters: int = None) -> Dict[int, List[IntelligentChunk]]:
"""Cluster chunks by semantic similarity."""
if not chunks or not any(chunk.semantic_embedding is not None for chunk in chunks):
return {0: chunks}
# Get embeddings
embeddings = []
valid_chunks = []
for chunk in chunks:
if chunk.semantic_embedding is not None:
embeddings.append(chunk.semantic_embedding)
valid_chunks.append(chunk)
if len(embeddings) < 2:
return {0: chunks}
embeddings = np.array(embeddings)
# Determine number of clusters
if n_clusters is None:
n_clusters = min(max(2, len(chunks) // 5), 10)
# Perform clustering
kmeans = KMeans(n_clusters=n_clusters, random_state=42)
cluster_labels = kmeans.fit_predict(embeddings)
# Group chunks by cluster
clusters = {}
for i, chunk in enumerate(valid_chunks):
cluster_id = int(cluster_labels[i])
if cluster_id not in clusters:
clusters[cluster_id] = []
clusters[cluster_id].append(chunk)
return clusters
def create_semantic_summary(self, chunks: List[IntelligentChunk]) -> Dict[str, Any]:
"""Create semantic summary of chunks."""
if not chunks:
return {}
# Aggregate metadata
content_types = {}
topics = {}
languages = {}
importance_scores = []
readability_scores = []
sentiment_scores = []
for chunk in chunks:
# Content types
ct = chunk.metadata.content_type
content_types[ct] = content_types.get(ct, 0) + 1
# Topics
topic = chunk.metadata.semantic_topic
topics[topic] = topics.get(topic, 0) + 1
# Languages
lang = chunk.metadata.language
languages[lang] = languages.get(lang, 0) + 1
# Scores
importance_scores.append(chunk.metadata.importance_score)
readability_scores.append(chunk.metadata.readability_score)
sentiment_scores.append(chunk.metadata.sentiment_score)
return {
'total_chunks': len(chunks),
'content_types': content_types,
'topics': topics,
'languages': languages,
'avg_importance': np.mean(importance_scores) if importance_scores else 0,
'avg_readability': np.mean(readability_scores) if readability_scores else 0,
'avg_sentiment': np.mean(sentiment_scores) if sentiment_scores else 0,
'total_entities': sum(chunk.metadata.entity_count for chunk in chunks)
}
def main():
"""Demo the intelligent chunking processor."""
print("๐Ÿง  Intelligent Chunking Processor Demo")
print("=" * 50)
# Initialize processor
processor = IntelligentChunkingProcessor()
# Demo content
demo_content = """
# Machine Learning Fundamentals
Machine learning is a subset of artificial intelligence that focuses on algorithms and statistical models.
## Key Concepts
### Supervised Learning
Supervised learning uses labeled training data to learn a mapping from inputs to outputs.
```python
from sklearn.linear_model import LinearRegression
model = LinearRegression()
model.fit(X_train, y_train)
predictions = model.predict(X_test)
```
### Unsupervised Learning
Unsupervised learning finds hidden patterns in data without labeled examples.
The K-means algorithm is a popular clustering method:
$$\\sum_{i=1}^{k} \\sum_{x \\in C_i} ||x - \\mu_i||^2$$
## Applications
Machine learning has numerous applications in:
- Computer vision
- Natural language processing
- Recommendation systems
- Autonomous vehicles
This technology is revolutionizing many industries and creating new opportunities.
"""
# Create intelligent chunks
print(f"\n๐Ÿ“ Processing content ({len(demo_content)} characters)...")
file_hash = hashlib.sha256(demo_content.encode()).hexdigest()
chunks = processor.create_intelligent_chunks(demo_content, file_hash)
print(f"โœ… Created {len(chunks)} intelligent chunks")
# Show chunk details
for i, chunk in enumerate(chunks):
print(f"\n๐Ÿ“„ Chunk {i+1}:")
print(f" Content type: {chunk.metadata.content_type}")
print(f" Topic: {chunk.metadata.semantic_topic}")
print(f" Importance: {chunk.metadata.importance_score:.2f}")
print(f" Readability: {chunk.metadata.readability_score:.2f}")
print(f" Entities: {chunk.metadata.entity_count}")
print(f" Language: {chunk.metadata.language}")
print(f" Content preview: {chunk.content[:100]}...")
# Create semantic summary
summary = processor.create_semantic_summary(chunks)
print(f"\n๐Ÿ“Š Semantic Summary:")
print(f" Total chunks: {summary['total_chunks']}")
print(f" Content types: {summary['content_types']}")
print(f" Topics: {summary['topics']}")
print(f" Average importance: {summary['avg_importance']:.2f}")
print(f" Average readability: {summary['avg_readability']:.2f}")
print(f" Total entities: {summary['total_entities']}")
print(f"\nโœ… Intelligent chunking processor ready!")
if __name__ == "__main__":
main()