File size: 5,345 Bytes
0dd9600
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# step3_encode_dataset_hybrid.py
import json
import os
import numpy as np
import pickle
from sentence_transformers import SentenceTransformer
from rank_bm25 import BM25Okapi
from underthesea import word_tokenize
import re

MODEL_NAME = "keepitreal/vietnamese-sbert"

INPUT_JSON = "dataset/knowledge_base.json"
OUT_DIR = "artifacts"
VECTORS_NPY = os.path.join(OUT_DIR, "kb_vectors.npy")
META_JSON = os.path.join(OUT_DIR, "kb_meta.json")
BM25_PKL = os.path.join(OUT_DIR, "bm25_index.pkl")
TOKENIZED_PKL = os.path.join(OUT_DIR, "tokenized_corpus.pkl")


def preprocess_vietnamese_text(text: str) -> str:
    """Chuẩn hóa text tiếng Việt"""
    if not text:
        return ""
    text = text.lower()
    # Giữ dấu tiếng Việt
    text = re.sub(r'[^\w\sàáạảãâầấậẩẫăằắặẳẵèéẹẻẽêềếệểễìíịỉĩòóọỏõôồốộổỗơờớợởỡùúụủũưừứựửữỳýỵỷỹđ]', ' ', text)
    text = ' '.join(text.split())
    return text


def extract_keywords(item: dict) -> list:
    keywords = []
    
    # Prioritize topic/chapter as exact phrases
    topic = item.get("topic", "").lower()
    if topic:
        keywords.append(topic)  # Keep full topic as one keyword
    
    chapter = item.get("chapter", "").lower()
    if chapter:
        keywords.append(chapter)
    
    # Add important bi-grams from content
    content = item.get("content_for_embedding", "").lower()
    if content:
        words = content.split()
        # Add 2-word phrases
        for i in range(len(words)-1):
            keywords.append(f"{words[i]} {words[i+1]}")
    
    return keywords[:30]  # Limit to avoid noise


def extract_text_for_embedding(item: dict) -> str:
    """Trích xuất text cho embedding"""
    texts = []
    
    topic = item.get("topic", "").strip()
    if topic:
        texts.append(f"Chủ đề: {topic}")
    
    content = item.get("content_for_embedding", "").strip()
    if content:
        texts.append(content)
    
    metadata = item.get("metadata", {})
    if isinstance(metadata, dict):
        raw_text = metadata.get("raw_text", "").strip()
        if raw_text and raw_text != content:
            texts.append(raw_text)
    
    chapter = item.get("chapter", "").strip()
    if chapter:
        texts.append(f"Thuộc: {chapter}")
    
    combined = ". ".join(texts)
    return preprocess_vietnamese_text(combined)


def main():
    os.makedirs(OUT_DIR, exist_ok=True)

    # Load dataset
    with open(INPUT_JSON, "r", encoding="utf-8") as f:
        data = json.load(f)

    print(f"📊 Processing {len(data)} items...")

    # Extract texts and metadata
    texts = []
    meta = []
    all_keywords = []
    
    for idx, item in enumerate(data):
        item_id = item.get("id", f"idx_{idx}")
        text = extract_text_for_embedding(item)
        keywords = extract_keywords(item)
        
        if not text or len(text) < 10:
            print(f"⚠️ Warning: Item {item_id} has insufficient text")
            continue
        
        texts.append(text)
        all_keywords.append(keywords)
        meta.append({
            "index": len(texts) - 1,
            "id": item_id,
            "topic": item.get("topic", ""),
            "chapter": item.get("chapter", ""),
            "knowledge_type": item.get("metadata", {}).get("knowledge_type", ""),
            "keywords": keywords,
            "text_length": len(text)
        })

    print(f"📏 Avg text length: {np.mean([m['text_length'] for m in meta]):.0f} chars")
    print(f"🔑 Avg keywords: {np.mean([len(k) for k in all_keywords]):.1f} per item")

    # ===== 1. Semantic Embeddings =====
    print(f"\n🤖 Loading model: {MODEL_NAME}")
    model = SentenceTransformer(MODEL_NAME)
    
    print("🔄 Encoding semantic vectors...")
    vectors = model.encode(
        texts,
        batch_size=32,
        show_progress_bar=True,
        normalize_embeddings=True,
        convert_to_numpy=True
    )
    vectors = np.asarray(vectors, dtype=np.float32)
    
    # ===== 2. BM25 Index =====
    print("\n📝 Building BM25 index...")
    tokenized_corpus = []
    
    for text in texts:
        try:
            # Tokenize tiếng Việt
            tokens = word_tokenize(text, format="text").split()
        except:
            # Fallback: simple split
            tokens = text.split()
        tokenized_corpus.append(tokens)
    
    bm25 = BM25Okapi(tokenized_corpus)
    
    # ===== 3. Save Everything =====
    print("\n💾 Saving artifacts...")
    
    np.save(VECTORS_NPY, vectors)
    
    with open(META_JSON, "w", encoding="utf-8") as f:
        json.dump(meta, f, ensure_ascii=False, indent=2)
    
    with open(BM25_PKL, "wb") as f:
        pickle.dump(bm25, f)
    
    with open(TOKENIZED_PKL, "wb") as f:
        pickle.dump(tokenized_corpus, f)

    print("\n✅ Step 3 DONE (Hybrid)")
    print(f"📦 Items: {len(texts)}")
    print(f"📐 Vector shape: {vectors.shape}")
    print(f"💾 Saved:")
    print(f"   - {VECTORS_NPY}")
    print(f"   - {META_JSON}")
    print(f"   - {BM25_PKL}")
    print(f"   - {TOKENIZED_PKL}")


if __name__ == "__main__":
    main()