File size: 5,910 Bytes
9381502
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
88c7275
 
 
 
 
 
 
 
 
 
 
 
 
 
9381502
 
88c7275
 
 
 
 
 
9381502
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from __future__ import annotations

import os
import re
from dataclasses import dataclass
from typing import Iterable, Optional

from cert_study_app.config import BASE_DIR


DEFAULT_EMBEDDING_MODEL = os.getenv("EMBEDDING_MODEL", "BAAI/bge-m3")


@dataclass
class VectorSearchResult:
    id: str
    text: str
    score: Optional[float] = None
    metadata: Optional[dict] = None


class QuestionVectorStore:
    def __init__(
        self,
        persist_dir: Optional[str] = None,
        collection_name: str = "questions",
        embedding_model: Optional[str] = None,
    ):
        self.persist_dir = persist_dir or str(BASE_DIR / "chroma_db")
        self.base_collection_name = collection_name
        self.embedding_model = embedding_model or DEFAULT_EMBEDDING_MODEL
        self.collection_name = _collection_name(collection_name, self.embedding_model)
        self._client = None
        self._collection = None

    def _get_collection(self):
        if self._collection is not None:
            return self._collection

        import chromadb
        from chromadb.utils import embedding_functions

        self._client = chromadb.PersistentClient(path=self.persist_dir)
        embedding_function = embedding_functions.SentenceTransformerEmbeddingFunction(
            model_name=self.embedding_model,
        )
        self._collection = self._client.get_or_create_collection(
            self.collection_name,
            embedding_function=embedding_function,
            metadata={
                "base_collection": self.base_collection_name,
                "embedding_model": self.embedding_model,
            },
        )
        return self._collection

    def upsert_questions(self, questions: Iterable[dict]) -> int:
        ids = []
        documents = []
        metadatas = []

        for item in questions:
            question_id = str(item["id"])
            ids.append(question_id)
            documents.append(_question_document(item))
            metadatas.append(
                {
                    "answer": item.get("answer", ""),
                    "category": item.get("category") or "",
                    "subcategory": item.get("subcategory") or "",
                    "source": item.get("source") or "",
                    "embedding_model": self.embedding_model,
                }
            )

        if not ids:
            return 0

        self._get_collection().upsert(ids=ids, documents=documents, metadatas=metadatas)
        return len(ids)

    def upsert_documents(self, documents: Iterable[dict]) -> int:
        ids = []
        texts = []
        metadatas = []

        for item in documents:
            doc_id = str(item["id"])
            text = str(item.get("text") or "").strip()
            if not doc_id or not text:
                continue
            ids.append(doc_id)
            texts.append(text)
            metadatas.append(
                {
                    "source_type": item.get("source_type") or self.base_collection_name,
                    "source": item.get("source") or "",
                    "title": item.get("title") or "",
                    "url": item.get("url") or "",
                    "category": item.get("category") or "",
                    "subcategory": item.get("subcategory") or "",
                    "embedding_model": self.embedding_model,
                }
            )

        if not ids:
            return 0

        self._get_collection().upsert(ids=ids, documents=texts, metadatas=metadatas)
        return len(ids)

    def search(self, query: str, k: int = 5, source: Optional[str] = None) -> list[VectorSearchResult]:
        try:
            collection = self._get_collection()
        except Exception:
            return []

        try:
            count = collection.count()
        except Exception:
            count = 0

        if count == 0:
            return []

        kwargs = {"query_texts": [query], "n_results": min(k, count)}
        if source:
            kwargs["where"] = {"source": source}

        try:
            result = collection.query(**kwargs)
        except Exception:
            return []

        ids = result.get("ids", [[]])[0]
        documents = result.get("documents", [[]])[0]
        distances = result.get("distances", [[]])[0] if result.get("distances") else []
        metadatas = result.get("metadatas", [[]])[0] if result.get("metadatas") else []

        items = []
        for idx, item_id in enumerate(ids):
            items.append(
                VectorSearchResult(
                    id=item_id,
                    text=documents[idx] if idx < len(documents) else "",
                    score=distances[idx] if idx < len(distances) else None,
                    metadata=metadatas[idx] if idx < len(metadatas) else None,
                )
            )
        return items


def _question_document(item: dict) -> str:
    parts = [f"문제: {item.get('stem') or ''}"]

    options = item.get("options") or {}
    if isinstance(options, dict) and options:
        rendered_options = "\n".join(f"{key}. {value}" for key, value in sorted(options.items()))
        parts.append(f"보기:\n{rendered_options}")
    elif isinstance(options, list) and options:
        parts.append("보기:\n" + "\n".join(str(option) for option in options))

    if item.get("answer"):
        parts.append(f"정답: {item['answer']}")
    if item.get("explanation"):
        parts.append(f"해설: {item['explanation']}")
    if item.get("category") or item.get("subcategory"):
        parts.append(f"분류: {item.get('category') or ''} / {item.get('subcategory') or ''}")

    return "\n".join(parts)


def _collection_name(base: str, embedding_model: str) -> str:
    slug = re.sub(r"[^a-zA-Z0-9._-]+", "_", embedding_model).strip("_").lower()
    name = f"{base}_{slug}" if slug else base
    return name[:63].strip("._-") or base