增加知识库，知识图谱，文件上传的字数限制

backend/api/chat_file.py

@@ -17,6 +17,7 @@ from core.database import get_db_pool
 from services.chat_thread_file_service import ChatThreadFileService
 from services.vector_service import get_vector_service
 from services.oss_service import get_oss_service
+from services.kb_text_limits import decode_txt_char_count, validate_chat_file_text_length
 from models.chat_thread_file import (
     ChatThreadFileUploadResponse,
     ChatThreadFileListResponse
@@ -438,6 +439,14 @@ async def upload_chat_file(
             )
         
         logger.info(f"✅ 文件大小验证通过: {file_size_mb:.2f}MB ({file_size} bytes)")
+
+        # .txt 文件上传时直接校验字符数，其他格式在后台处理时校验
+        if file_ext == ".txt":
+            try:
+                validate_chat_file_text_length(decode_txt_char_count(content))
+            except ValueError as e:
+                logger.warning(f"❌ 文本文件字数超限: {file.filename}, {e}")
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=str(e))
         
         # 生成唯一文件名（使用时间戳）
         timestamp = int(time.time() * 1000)

backend/api/kb_file_router.py

@@ -23,6 +23,7 @@ from services.knowledge_base_service import KnowledgeBaseService
 from services.knowledge_base_file_service import KnowledgeBaseFileService
 from services.audit_service import AuditService
 from services.vector_service import get_vector_service
+from services.kb_text_limits import decode_txt_char_count, validate_kb_text_length
 from services.oss_service import get_oss_service
 from utils.helpers import BaseResponse
 from logger.logging import get_logger
@@ -356,6 +357,13 @@ async def upload_file(
             raise BadRequestError(f"文件大小超过限制，当前: {file_size_mb:.2f}MB，最大允许: 15MB")
         
         logger.info(f"✅ 文件大小验证通过: {file_size_mb:.2f}MB ({file_size} bytes)")
+
+        if file_ext == ".txt":
+            try:
+                validate_kb_text_length(decode_txt_char_count(content))
+            except ValueError as e:
+                logger.warning(f"❌ 文本文件字数超限: {file.filename}, {e}")
+                raise BadRequestError(str(e)) from e
         
         # 生成唯一文件名
         timestamp = int(time.time() * 1000)

backend/api/knowledge_graph_router.py

@@ -22,6 +22,7 @@ from services import neo4j_service
 from services.novel_kg_service import (
     extract_and_import_knowledge_graph,
     extract_knowledge_document_text,
+    validate_knowledge_graph_text_length,
 )
 from utils.helpers import BaseResponse
 from logger.logging import get_logger
@@ -139,6 +140,7 @@ async def create_knowledge_graph(
 
     try:
         text = await extract_knowledge_document_text(file.filename, raw)
+        validate_knowledge_graph_text_length(text)
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e)) from e
 

backend/services/kb_text_limits.py

@@ -0,0 +1,57 @@
+"""
+知识库与对话文件的正文长度限制。
+
+处理流程对比：
+  知识图谱  每块 1 次串行 LLM 抽关系 → 最严格，8 万字
+  知识库    批量 embedding + 单次摘要 → 较宽松，30 万字
+  对话文件  批量 embedding + 单次摘要（临时会话上下文）→ 20 万字
+"""
+
+# ---------- 知识库 ----------
+# chunk_size=4096、overlap=200 → 有效步长约 3896 字/块
+# 30 万字 ≈ 77 块向量化，通常 1-2 分钟可完成
+MAX_KB_INPUT_CHARS = 300_000
+MAX_KB_VECTOR_CHUNKS = 80
+
+# ---------- 对话文件（聊天时上传） ----------
+# 对话文件是单次会话临时上下文，精度要求高于知识库；过长会稀释检索精度
+# 20 万字 ≈ 51 块，兼顾处理速度与上下文覆盖
+MAX_CHAT_FILE_INPUT_CHARS = 200_000
+MAX_CHAT_FILE_VECTOR_CHUNKS = 60
+
+
+def decode_txt_char_count(raw: bytes) -> int:
+    """估算纯文本文件字符数（上传前快速校验）。"""
+    try:
+        text = raw.decode("utf-8")
+    except UnicodeDecodeError:
+        text = raw.decode("gb18030", errors="replace")
+    return len(text.strip())
+
+
+def validate_kb_text_length(char_count: int, *, chunk_count: int | None = None) -> None:
+    """校验知识库可处理的正文规模。"""
+    if char_count > MAX_KB_INPUT_CHARS:
+        raise ValueError(
+            f"提取的正文过长（约 {char_count:,} 字），知识库单文件上限为 {MAX_KB_INPUT_CHARS:,} 字。"
+            "请将文档拆分为多个文件分别上传。"
+        )
+    if chunk_count is not None and chunk_count > MAX_KB_VECTOR_CHUNKS:
+        raise ValueError(
+            f"文本分块过多（{chunk_count} 块，上限 {MAX_KB_VECTOR_CHUNKS} 块），"
+            f"请将正文控制在约 {MAX_KB_INPUT_CHARS:,} 字以内后重试。"
+        )
+
+
+def validate_chat_file_text_length(char_count: int, *, chunk_count: int | None = None) -> None:
+    """校验对话文件可处理的正文规模。"""
+    if char_count > MAX_CHAT_FILE_INPUT_CHARS:
+        raise ValueError(
+            f"提取的正文过长（约 {char_count:,} 字），对话文件上限为 {MAX_CHAT_FILE_INPUT_CHARS:,} 字。"
+            "请将文档拆分后上传，或使用知识库功能处理长文档。"
+        )
+    if chunk_count is not None and chunk_count > MAX_CHAT_FILE_VECTOR_CHUNKS:
+        raise ValueError(
+            f"文本分块过多（{chunk_count} 块，上限 {MAX_CHAT_FILE_VECTOR_CHUNKS} 块），"
+            f"请将正文控制在约 {MAX_CHAT_FILE_INPUT_CHARS:,} 字以内后重试。"
+        )

backend/services/novel_kg_service.py

@@ -23,7 +23,10 @@ from logger.logging import get_logger
 
 logger = get_logger(__name__)
 
-MAX_INPUT_CHARS = 800_000
+# 知识图谱抽取上限：每块约 900 字、重叠 120，每块 1 次 DeepSeek 调用（串行）。
+# 8 万字 ≈ 100 次调用，后台约 5–10 分钟可完成；50 万字需 600+ 次，基本不可行。
+MAX_INPUT_CHARS = 80_000
+MAX_KG_EXTRACT_CHUNKS = 100
 CHUNK_SIZE = 900
 CHUNK_OVERLAP = 120
 
@@ -644,17 +647,32 @@ def merge_triplets(chunks: list[list[dict[str, Any]]]) -> list[dict[str, Any]]:
     return merged
 
 
+def validate_knowledge_graph_text_length(text: str) -> None:
+    """上传/构建前校验正文长度，避免超长文本进入图谱抽取流水线。"""
+    char_count = len(text)
+    if char_count > MAX_INPUT_CHARS:
+        raise ValueError(
+            f"提取的正文过长（约 {char_count:,} 字），知识图谱构建上限为 {MAX_INPUT_CHARS:,} 字。"
+            "请拆分为多个文件分别上传，或使用知识库功能处理长文档问答。"
+        )
+
+
 async def extract_and_import_knowledge_graph(text: str, graph_id: str) -> dict[str, Any]:
     """
     对整篇文本分块调用 LLM，合并三元组后写入 Neo4j。
     """
-    if len(text) > MAX_INPUT_CHARS:
+    validate_knowledge_graph_text_length(text)
-        raise ValueError(f"文本过长，请控制在约 {MAX_INPUT_CHARS} 字以内")
 
     chunks = split_novel_text(text)
     if not chunks:
         raise ValueError("文本为空")
 
+    if len(chunks) > MAX_KG_EXTRACT_CHUNKS:
+        raise ValueError(
+            f"文本分块过多（{len(chunks)} 块，上限 {MAX_KG_EXTRACT_CHUNKS} 块），"
+            f"请将正文控制在约 {MAX_INPUT_CHARS:,} 字以内后重试。"
+        )
+
     logger.info("知识图谱：共 {} 个文本块", len(chunks))
     batch_results: list[list[dict[str, Any]]] = []
     for i, ch in enumerate(chunks):

backend/services/vector_service.py

@@ -78,6 +78,7 @@ import bs4
 
 from logger.logging import get_logger
 from core.config import settings
+from services.kb_text_limits import validate_kb_text_length, validate_chat_file_text_length
 
 logger = get_logger(__name__)
 
@@ -859,6 +860,14 @@ class VectorService:
                 error_msg = "未能从文件加载到任何内容"
                 logger.warning(error_msg)
                 return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
+
+            total_chars = sum(len(d.page_content or "") for d in docs)
+            try:
+                validate_kb_text_length(total_chars)
+            except ValueError as e:
+                error_msg = str(e)
+                logger.warning(error_msg)
+                return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
             
             # 2. 分割文本
             all_splits = self.text_splitter.split_documents(docs)
@@ -869,6 +878,13 @@ class VectorService:
                 error_msg = "文档分割后没有内容，可能是空白文档"
                 logger.warning(error_msg)
                 return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
+
+            try:
+                validate_kb_text_length(total_chars, chunk_count=len(all_splits))
+            except ValueError as e:
+                error_msg = str(e)
+                logger.warning(error_msg)
+                return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
             
             # 3. 向量化并存储
             collection_name = f"kb_{knowledge_base_id}"
@@ -961,6 +977,14 @@ class VectorService:
                 error_msg = "未能从 URL 加载到任何内容"
                 logger.warning(error_msg)
                 return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
+
+            total_chars = sum(len(d.page_content or "") for d in docs)
+            try:
+                validate_kb_text_length(total_chars)
+            except ValueError as e:
+                error_msg = str(e)
+                logger.warning(error_msg)
+                return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
             
             # 2. 分割文本
             all_splits = self.text_splitter.split_documents(docs)
@@ -971,6 +995,13 @@ class VectorService:
                 error_msg = "网页分割后没有内容，可能是空白页面或无法提取文本"
                 logger.warning(error_msg)
                 return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
+
+            try:
+                validate_kb_text_length(total_chars, chunk_count=len(all_splits))
+            except ValueError as e:
+                error_msg = str(e)
+                logger.warning(error_msg)
+                return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
             
             # 3. 向量化并存储
             collection_name = f"kb_{knowledge_base_id}"
@@ -1081,6 +1112,14 @@ class VectorService:
                 error_msg = "未能加载到任何内容"
                 logger.warning(error_msg)
                 return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
+
+            total_chars = sum(len(d.page_content or "") for d in docs)
+            try:
+                validate_chat_file_text_length(total_chars)
+            except ValueError as e:
+                error_msg = str(e)
+                logger.warning(error_msg)
+                return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
             
             # 分割文本
             all_splits = self.text_splitter.split_documents(docs)
@@ -1091,6 +1130,13 @@ class VectorService:
                 error_msg = "文档分割后没有内容，可能是空白文档或无法提取文本"
                 logger.warning(error_msg)
                 return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
+
+            try:
+                validate_chat_file_text_length(total_chars, chunk_count=len(all_splits))
+            except ValueError as e:
+                error_msg = str(e)
+                logger.warning(error_msg)
+                return ProcessResult(success=False, chunks=[], chunk_count=0, error_message=error_msg)
             
             # 向量化并存储（使用 thread_id 作为集合名）
             collection_name = f"thread_{thread_id}"

frontend/src/views/KnowledgeBase.vue

@@ -216,7 +216,7 @@
                       ref="fileInput"
                     >
                   </label>
-                <small class="upload-hint">支持 PDF、DOCX、Excel（xlsx/xls）、CSV、TXT、图片（PNG、JPG、BMP）格式，文件将自动进行向量化处理</small>
+                <small class="upload-hint">支持 PDF、DOCX、Excel（xlsx/xls）、CSV、TXT、图片（PNG、JPG、BMP）格式，单文件不超过 15MB，提取正文不超过 30 万字（过长请拆分）</small>
                 </div>
 
                 <!-- URL 上传 -->

frontend/src/views/KnowledgeGraph.vue

@@ -219,8 +219,7 @@
               class="form-control bg-dark text-white border-secondary"
               @change="onFileSelect"
             />
-            <div class="mt-2 small text-muted">支持 .txt、.pdf、.docx 及常见图片；扫描件将尝试 OCR + 通义视觉。单文件不超过约 15MB</div>
+            <div class="mt-2 small text-muted">支持 .txt、.pdf、.docx 及常见图片；扫描件将尝试 OCR + 通义视觉。单文件不超过约 15MB，提取正文建议不超过 8 万字（过长请拆分或改用知识库）</div>          </div>
-          </div>
           <div v-if="uploadError" class="alert alert-danger py-2 small">{{ uploadError }}</div>
         </div>
         <div class="modal-footer-custom">