Spaces:
Sleeping
Sleeping
File size: 9,456 Bytes
300df0f | 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 | import pandas as pd
import os
import logging
import re
import json
from src.segmentation.parser import LegalDocumentParser
from src.segmentation.writer import SegmentWriter
from src.segmentation.confidence import ConfidenceScorer
from src.cross_reference.extractor import CrossReferenceExtractor
logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s — %(message)s")
logger = logging.getLogger(__name__)
NEO4J_URI = os.getenv("NEO4J_URI", "bolt://localhost:7687")
NEO4J_USER = os.getenv("NEO4J_USER", "neo4j")
NEO4J_PASS = os.getenv("NEO4J_PASSWORD", "password")
def run_unified_pipeline():
# 1.1 Infrastructure Skeleton
scorer = ConfidenceScorer()
parser = LegalDocumentParser()
logger.info("Đọc metadata_deduped.parquet...")
meta_df = pd.read_parquet("data/metadata_deduped.parquet")
meta_df['id'] = meta_df['id'].astype(str)
# 1.2 Filter loai_van_ban and ngay_ban_hanh
core_types = ['Thông tư', 'Nghị định', 'Luật', 'Bộ luật']
validity = ['Còn hiệu lực','Hết hiệu lực một phần']
meta_df['ngay_ban_hanh'] = pd.to_datetime(meta_df['ngay_ban_hanh'], errors='coerce', dayfirst=True)
filtered_meta = meta_df[
(meta_df['loai_van_ban'].isin(core_types)) &
(meta_df['tinh_trang_hieu_luc'].isin(validity)) &
(meta_df['ngay_ban_hanh'] >= '2000-01-01')
]
logger.info(f"Tổng số văn bản sau khi lọc: {len(filtered_meta)}")
# Load lookup table from JSON (Centralized for all scripts)
lookup_path = "data/so_ky_hieu_lookup.json"
if not os.path.exists(lookup_path):
logger.info("Không thấy bảng tra cứu JSON, đang tạo mới...")
from src.data_pipeline.build_lookup_json import build
build()
with open(lookup_path, "r", encoding="utf-8") as f:
lookup = json.load(f)
logger.info(f"Đã nạp bảng tra cứu từ JSON ({len(lookup)} mục).")
extractor = CrossReferenceExtractor(lookup_table=lookup)
logger.info("Đọc content_clean.parquet...")
content_df = pd.read_parquet("data/content_clean.parquet")
content_df['id'] = content_df['id'].astype(str)
content_dict = content_df.set_index('id')['clean_html'].to_dict()
total_docs = 0
batch_size = 50 # 3.1 Batching to prevent memory issues
batch_segments = []
part_idx = 0
# Lists for relationships
internal_refs_data = []
external_refs_data = []
modifies_refs_data = []
doc_id = '153913'
meta_row = filtered_meta[filtered_meta['id'] == doc_id]
if meta_row.empty:
logger.error(f"Văn bản ID {doc_id} không tìm thấy trong metadata.")
return
row = meta_row.iloc[0]
html = content_dict.get(doc_id, "")
try:
# 2.1 Stage 2: Preamble Extraction
logger.info(f"Start stage 2: Preamble Extraction...")
preamble_text = ""
parts = re.split(r'(<[^>]+>\s*Điều\s+1[\.:\s])', html, maxsplit=1, flags=re.IGNORECASE)
if len(parts) > 1:
preamble_text = parts[0]
primary_target_ref = extractor._extract_preamble_anchor(preamble_text)
if primary_target_ref:
logger.info(f"Tìm thấy văn bản đích từ lời nói đầu: {primary_target_ref.raw_so_ky_hieu} (ID: {primary_target_ref.target_doc_id})")
# 2.2 Stage 3: Segmentation
logger.info(f"Start stage 3: Segmentation...")
result = parser.parse(
doc_id=doc_id,
clean_html=html,
loai_van_ban=row.get('loai_van_ban', '')
)
logger.info(f"Phân tích segmentation: {len(result.segments)} segments")
# 2.3 & 2.4 Stage 4: Cross-Reference & Context-Aware Extraction
logger.info(f"Start stage 4: Cross-Reference & Context-Aware Extraction...")
is_modifying = (primary_target_ref is not None) or ("sửa đổi" in str(row.get('title', '')).lower())
from src.segmentation.models import HierarchyType
all_relationships = []
def parse_uid_parts(uid):
if not uid: return "", "", ""
p = uid.split('_')
d = k = di = ""
if 'dieu' in p: d = p[p.index('dieu')+1]
if 'khoan' in p: k = p[p.index('khoan')+1]
if 'diem' in p: di = p[p.index('diem')+1]
return d, k, di
logger.info(f"Executing stage 4: Parse done...")
uid_to_seg = {s.uid: s for s in result.segments if s.uid}
last_target_doc_id = None
last_target_article = None
for seg in result.segments:
if seg.hierarchy_type not in [HierarchyType.DIEU, HierarchyType.KHOAN, HierarchyType.DIEM]:
continue
# Xác định article_uid, clause_uid, point_uid cho source
art_uid = cl_uid = pt_uid = None
curr = seg
if curr.hierarchy_type == HierarchyType.DIEM:
pt_uid = curr.uid
curr = uid_to_seg.get(curr.parent_uid)
if curr and curr.hierarchy_type == HierarchyType.KHOAN:
cl_uid = curr.uid
curr = uid_to_seg.get(curr.parent_uid)
if curr and curr.hierarchy_type == HierarchyType.DIEU:
art_uid = curr.uid
if not art_uid: continue
ext_result = extractor.extract_from_article(
doc_id=doc_id,
article_uid=art_uid,
clause_uid=cl_uid,
point_uid=pt_uid,
article_text=seg.clean_text,
is_modifying_doc=is_modifying
)
src_art, src_cl, src_pt = parse_uid_parts(seg.uid)
# Thu thập Internal Refs
for r in ext_result.internal_refs:
all_relationships.append({
"src_doc": doc_id, "src_art": src_art, "src_cl": src_cl, "src_pt": src_pt,
"tgt_doc": doc_id, "tgt_art": r.target_article_index or "",
"tgt_cl": r.target_clause_index or "", "tgt_pt": r.target_point_label or "",
"type": "Internal", "context": r.context_text.replace('\n', ' ')
})
# Thu thập External Refs
for r in ext_result.external_refs:
all_relationships.append({
"src_doc": doc_id, "src_art": src_art, "src_cl": src_cl, "src_pt": src_pt,
"tgt_doc": r.target_doc_id or r.raw_so_ky_hieu, "tgt_art": r.target_article_index or "",
"tgt_cl": r.target_clause_index or "", "tgt_pt": r.target_point_label or "",
"type": "External", "context": r.context_text.replace('\n', ' ')
})
# Thu thập Modification Refs
for r in ext_result.modification_refs:
# Bổ sung thông tin target_doc_id nếu thiếu từ primary target (lời nói đầu)
if not r.target_doc_id and primary_target_ref:
r.target_doc_id = primary_target_ref.target_doc_id
# Logic ROLL BACK: Nếu ref thiếu Điều đích, lấy từ ref trước đó
if r.is_partial_ref and last_target_article:
r.target_article_index = last_target_article
if not r.target_doc_id:
r.target_doc_id = last_target_doc_id
# Cập nhật state cho các ref tiếp theo
if r.target_article_index:
last_target_article = r.target_article_index
last_target_doc_id = r.target_doc_id
all_relationships.append({
"src_doc": doc_id, "src_art": src_art, "src_cl": src_cl, "src_pt": src_pt,
"tgt_doc": r.target_doc_id or r.raw_target_so_ky_hieu, "tgt_art": r.target_article_index or "",
"tgt_cl": r.target_clause_index or "", "tgt_pt": r.target_point_label or "",
"type": f"Modification ({r.action.value if hasattr(r.action, 'value') else r.action})",
"context": r.context_text.replace('\n', ' ')
})
logger.info(f"Executing stage 4: Extract done...")
# Xuất file Markdown
output_file = f"test_{doc_id}.md"
with open(output_file, "w", encoding="utf-8") as f:
f.write(f"# Kết quả trích dẫn quan hệ - Văn bản {doc_id}\n\n")
f.write("| Source Docs | Article | Clause | Point | Target Docs | Article | Clause | Point | Type | Context |\n")
f.write("|-------------|---------|--------|-------|-------------|---------|--------|-------|------|---------|\n")
for rel in all_relationships:
f.write(f"| {rel['src_doc']} | {rel['src_art']} | {rel['src_cl']} | {rel['src_pt']} | "
f"{rel['tgt_doc']} | {rel['tgt_art']} | {rel['tgt_cl']} | {rel['tgt_pt']} | "
f"{rel['type']} | {rel['context']} |\n")
logger.info(f"Đã xuất kết quả ra file: {output_file}")
except Exception as e:
logger.error(f"Lỗi xử lý nội dung văn bản {doc_id}: {e}")
import traceback
logger.error(traceback.format_exc())
logger.info("=== HOÀN TẤT UNIFIED PIPELINE ===")
if __name__ == "__main__":
run_unified_pipeline()
|