File size: 8,127 Bytes
c0f31c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""

SemanticChunker.py

A module for semantic-aware text chunking using embeddings and similarity metrics.

"""

import logging
from typing import List, Optional, Any
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
from langchain_core.documents import Document
# FIXED IMPORT: Updated for LangChain v0.2+
from langchain_text_splitters import SpacyTextSplitter 
from sentence_transformers import SentenceTransformer
from core.BaseChunker import BaseChunker

logger = logging.getLogger(__name__)

class SemanticChunker(BaseChunker):
    """Chunks text based on semantic similarity and size constraints"""
    
    def __init__(

        self, 

        model_name: Optional[str] = None,

        embedding_model: Optional[Any] = None,

        chunk_size: int = 200, 

        chunk_overlap: int = 0, 

        similarity_threshold: float = 0.9, 

        separator: str = " "

    ):
        """

        Initialize the semantic chunker with configurable parameters

        """
        # Validate parameters
        if chunk_size <= 0:
            raise ValueError("chunk_size must be a positive integer.")
        if not (0 <= similarity_threshold <= 1):
            raise ValueError("similarity_threshold must be between 0 and 1.")
            
        # Initialize BaseChunker first
        super().__init__(model_name, embedding_model)
        
        # Set semantic chunking parameters
        self.chunk_size = chunk_size
        self.chunk_overlap = chunk_overlap
        self.similarity_threshold = similarity_threshold
        self.separator = separator
        
        # Use provided embedding model or initialize sentence transformer
        is_dummy = False
        if embedding_model is not None:
            try:
                test_output = embedding_model.encode("test")
                if isinstance(test_output, list) and len(test_output) == 384 and all(x == 0.0 for x in test_output):
                    is_dummy = True
            except:
                pass

        if embedding_model is None or is_dummy:
            try:
                self.sentence_model = SentenceTransformer('multi-qa-mpnet-base-dot-v1')
                self.embedding_model = self.sentence_model
                logger.info("Initialized SentenceTransformer for semantic chunking")
            except Exception as e:
                logger.error(f"Error loading SentenceTransformer: {e}")
                class DummyEmbedder:
                    def encode(self, text, **kwargs):
                        return [0.0] * 384
                self.sentence_model = DummyEmbedder()
                self.embedding_model = self.sentence_model
        else:
            self.sentence_model = embedding_model
            logger.info("Using provided embedding model for semantic chunking")
        
        # Initialize text splitter for initial chunking
        self.text_splitter = SpacyTextSplitter(
            chunk_size=self.chunk_size - self.chunk_overlap,
            chunk_overlap=self.chunk_overlap,
            separator=self.separator 
        )

    def _enforce_size_immediately(self, text: str) -> List[str]:
        if not text.strip():
            return []
            
        chunks = []
        current_chunk = []
        words = text.split()
        
        for word in words:
            if sum(len(w) for w in current_chunk) + len(word) + len(current_chunk) <= self.chunk_size:
                current_chunk.append(word)
            else:
                if current_chunk:
                    chunks.append(" ".join(current_chunk))
                current_chunk = [word]
                
        if current_chunk:
            chunks.append(" ".join(current_chunk))
            
        return chunks

    def get_semantic_chunks(self, documents: List[Document]) -> List[Document]:
        if not documents:
            logger.warning("No documents provided for semantic chunking")
            return []
            
        try:
            base_chunks = self.text_splitter.split_documents(documents)
            logger.info(f"Initial splitting created {len(base_chunks)} base chunks")
            
            if not base_chunks:
                return []
                
            chunk_contents = [doc.page_content for doc in base_chunks]
            chunk_embeddings = self.sentence_model.encode(chunk_contents)
            
            grouped_chunks = []
            current_group = []
            current_embedding = None

            for i, base_chunk in enumerate(base_chunks):
                if not current_group:
                    current_group.append(base_chunk)
                    current_embedding = chunk_embeddings[i].reshape(1, -1)
                    continue
                    
                similarity = cosine_similarity(current_embedding, chunk_embeddings[i].reshape(1, -1))[0][0]
                combined_content = " ".join([doc.page_content for doc in current_group] + [base_chunk.page_content])

                if similarity >= self.similarity_threshold and len(combined_content) <= self.chunk_size:
                    current_group.append(base_chunk)
                else:
                    grouped_chunks.extend(self._finalize_chunk_group(current_group))
                    current_group = [base_chunk]
                    current_embedding = chunk_embeddings[i].reshape(1, -1)
                    
            if current_group:
                grouped_chunks.extend(self._finalize_chunk_group(current_group))
                
            logger.info(f"Created {len(grouped_chunks)} semantic chunks")
            return grouped_chunks
            
        except Exception as e:
            logger.error(f"Error in semantic chunking: {e}")
            return documents

    def _finalize_chunk_group(self, group: List[Document]) -> List[Document]:
        if not group:
            return []
            
        processed_chunks = []
        content = " ".join([doc.page_content for doc in group])
        size_limited_chunks = self._enforce_size_immediately(content)
        
        base_metadata = group[0].metadata.copy()
        
        for i, chunk in enumerate(size_limited_chunks):
            stats = self.analyze_text(chunk)
            metadata = base_metadata.copy()
            metadata.update({
                "chunk_index": i + 1,
                "chunk_count": len(size_limited_chunks),
                "char_count": stats["char_count"],
                "token_count": stats["token_count"],
                "sentence_count": stats["sentence_count"],
                "word_count": stats["word_count"],
                "chunk_type": "semantic"
            })
            
            processed_chunks.append(Document(page_content=chunk, metadata=metadata))
            
        return processed_chunks

    def semantic_process_document(self, file_path: str, preprocess: bool = False) -> List[Document]:
        try:
            logger.info(f"Processing document with semantic chunking: {file_path}")
            
            raw_documents = self.load_document(file_path)
            
            processed_documents = []
            for doc in raw_documents:
                content = doc.page_content
                if preprocess:
                    content = self.preprocess_text(content)
                processed_documents.append(Document(
                    page_content=content,
                    metadata=doc.metadata
                ))
            
            documents = self.get_semantic_chunks(processed_documents)
            logger.info(f"Created {len(documents)} semantic chunks")
            
            return documents
            
        except Exception as e:
            logger.error(f"Error in semantic_process_document: {e}")
            raise
    
    def process_document(self, file_path: str, preprocess: bool = True) -> List[Document]:
        return self.semantic_process_document(file_path, preprocess)