feat: 生成关联问题

--story=1016225 --user=刘瑞斌 【北区】在现有的分段基础上，加个 「生成问题」 ，支持调用大模型 生成问题并关联到分段。 https://www.tapd.cn/57709429/s/1589572

apps/dataset/models/data_set.py

@@ -22,6 +22,7 @@ class Status(models.TextChoices):
     success = 1, '已完成'
     error = 2, '导入失败'
     queue_up = 3, '排队中'
+    generating = 4, '生成问题中'
 
 
 class Type(models.TextChoices):

apps/dataset/serializers/document_serializers.py

@@ -47,7 +47,7 @@ from dataset.models.data_set import DataSet, Document, Paragraph, Problem, Type,
 from dataset.serializers.common_serializers import BatchSerializer, MetaSerializer, ProblemParagraphManage, \
     get_embedding_model_id_by_dataset_id
 from dataset.serializers.paragraph_serializers import ParagraphSerializers, ParagraphInstanceSerializer
-from dataset.task import sync_web_document
+from dataset.task import sync_web_document, generate_related_by_document_id
 from embedding.task.embedding import embedding_by_document, delete_embedding_by_document_list, \
     delete_embedding_by_document, update_embedding_dataset_id, delete_embedding_by_paragraph_ids, \
     embedding_by_document_list
@@ -960,6 +960,37 @@ class DocumentSerializers(ApiMixin, serializers.Serializer):
                     except AlreadyQueued as e:
                         raise AppApiException(500, "任务正在执行中,请勿重复下发")
 
+    class GenerateRelated(ApiMixin, serializers.Serializer):
+        document_id = serializers.UUIDField(required=True, error_messages=ErrMessage.uuid("文档id"))
+
+        def is_valid(self, *, raise_exception=False):
+            super().is_valid(raise_exception=True)
+            document_id = self.data.get('document_id')
+            if not QuerySet(Document).filter(id=document_id).exists():
+                raise AppApiException(500, "文档id不存在")
+
+        def generate_related(self, model_id, prompt, with_valid=True):
+            if with_valid:
+                self.is_valid(raise_exception=True)
+            document_id = self.data.get('document_id')
+            QuerySet(Document).filter(id=document_id).update(status=Status.queue_up)
+            generate_related_by_document_id.delay(document_id, model_id, prompt)
+
+
+
+    class BatchGenerateRelated(ApiMixin, serializers.Serializer):
+        dataset_id = serializers.UUIDField(required=True, error_messages=ErrMessage.uuid("知识库id"))
+
+        @transaction.atomic
+        def batch_generate_related(self, instance: Dict, with_valid=True):
+            if with_valid:
+                self.is_valid(raise_exception=True)
+            document_id_list = instance.get("document_id_list")
+            model_id = instance.get("model_id")
+            prompt = instance.get("prompt")
+            for document_id in document_id_list:
+                DocumentSerializers.GenerateRelated(data={'document_id': document_id}).generate_related(model_id, prompt)
+
 
 class FileBufferHandle:
     buffer = None

apps/dataset/serializers/paragraph_serializers.py

@@ -27,6 +27,7 @@ from embedding.models import SourceType
 from embedding.task.embedding import embedding_by_problem as embedding_by_problem_task, embedding_by_problem, \
     delete_embedding_by_source, enable_embedding_by_paragraph, disable_embedding_by_paragraph, embedding_by_paragraph, \
     delete_embedding_by_paragraph, delete_embedding_by_paragraph_ids, update_embedding_document_id
+from dataset.task import generate_related_by_paragraph_id_list
 
 
 class ParagraphSerializer(serializers.ModelSerializer):
@@ -719,3 +720,20 @@ class ParagraphSerializers(ApiMixin, serializers.Serializer):
                                                   )
                 }
             )
+
+
+    class BatchGenerateRelated(ApiMixin, serializers.Serializer):
+        dataset_id = serializers.UUIDField(required=True, error_messages=ErrMessage.uuid("知识库id"))
+        document_id = serializers.UUIDField(required=True, error_messages=ErrMessage.uuid("文档id"))
+
+        @transaction.atomic
+        def batch_generate_related(self, instance: Dict, with_valid=True):
+            if with_valid:
+                self.is_valid(raise_exception=True)
+            paragraph_id_list = instance.get("paragraph_id_list")
+            model_id = instance.get("model_id")
+            prompt = instance.get("prompt")
+            generate_related_by_paragraph_id_list.delay(paragraph_id_list, model_id, prompt)
+
+
+

apps/dataset/task/__init__.py

@@ -7,3 +7,4 @@
     @desc:
 """
 from .sync import *
+from .generate import *

apps/dataset/task/generate.py

@@ -0,0 +1,64 @@
+import logging
+from math import ceil
+
+from celery_once import QueueOnce
+from django.db.models import QuerySet
+from langchain_core.messages import HumanMessage
+
+from common.config.embedding_config import ModelManage
+from dataset.models import Paragraph, Document, Status
+from dataset.task.tools import save_problem
+from ops import celery_app
+from setting.models import Model
+from setting.models_provider import get_model
+
+max_kb_error = logging.getLogger("max_kb_error")
+max_kb = logging.getLogger("max_kb")
+
+
+def get_llm_model(model_id):
+    model = QuerySet(Model).filter(id=model_id).first()
+    return ModelManage.get_model(model_id, lambda _id: get_model(model))
+
+
+@celery_app.task(base=QueueOnce, once={'keys': ['document_id']},
+                 name='celery:generate_related_by_document')
+def generate_related_by_document_id(document_id, model_id, prompt):
+    llm_model = get_llm_model(model_id)
+    offset = 0
+    page_size = 10
+    QuerySet(Document).filter(id=document_id).update(status=Status.generating)
+
+    count = QuerySet(Paragraph).filter(document_id=document_id).count()
+    for i in range(0, ceil(count / page_size)):
+        paragraph_list = QuerySet(Paragraph).filter(document_id=document_id).all()[offset:offset + page_size]
+        offset += page_size
+        for paragraph in paragraph_list:
+            res = llm_model.invoke([HumanMessage(content=prompt.replace('{data}', paragraph.content))])
+            if (res.content is None) or (len(res.content) == 0):
+                continue
+            problems = res.content.split('\n')
+            for problem in problems:
+                save_problem(paragraph.dataset_id, paragraph.document_id, paragraph.id, problem)
+
+    QuerySet(Document).filter(id=document_id).update(status=Status.success)
+
+
+
+@celery_app.task(base=QueueOnce, once={'keys': ['paragraph_id_list']},
+                 name='celery:generate_related_by_paragraph_list')
+def generate_related_by_paragraph_id_list(paragraph_id_list, model_id, prompt):
+    llm_model = get_llm_model(model_id)
+    offset = 0
+    page_size = 10
+    count = QuerySet(Paragraph).filter(id__in=paragraph_id_list).count()
+    for i in range(0, ceil(count / page_size)):
+        paragraph_list = QuerySet(Paragraph).filter(id__in=paragraph_id_list).all()[offset:offset + page_size]
+        offset += page_size
+        for paragraph in paragraph_list:
+            res = llm_model.invoke([HumanMessage(content=prompt.replace('{data}', paragraph.content))])
+            if (res.content is None) or (len(res.content) == 0):
+                continue
+            problems = res.content.split('\n')
+            for problem in problems:
+                save_problem(paragraph.dataset_id, paragraph.document_id, paragraph.id, problem)

apps/dataset/task/tools.py

@@ -8,6 +8,7 @@
 """
 
 import logging
+import re
 import traceback
 
 from common.util.fork import ChildLink, Fork
@@ -60,3 +61,23 @@ def get_sync_web_document_handler(dataset_id):
                      status=Status.error).save()
 
     return handler
+
+
+def save_problem(dataset_id, document_id, paragraph_id, problem):
+    from dataset.serializers.paragraph_serializers import ParagraphSerializers
+    # print(f"dataset_id: {dataset_id}")
+    # print(f"document_id: {document_id}")
+    # print(f"paragraph_id: {paragraph_id}")
+    # print(f"problem: {problem}")
+    problem = re.sub(r"^\d+\.\s*", "", problem)
+    pattern = r"<question>(.*?)</question>"
+    match = re.search(pattern, problem)
+    problem = match.group(1) if match else None
+    if problem is None or len(problem) == 0:
+        return
+    try:
+        ParagraphSerializers.Problem(
+            data={"dataset_id": dataset_id, 'document_id': document_id,
+                  'paragraph_id': paragraph_id}).save(instance={"content": problem}, with_valid=True)
+    except Exception as e:
+        max_kb_error.error(f'关联问题失败: {e}')

apps/dataset/urls.py

@@ -15,6 +15,7 @@ urlpatterns = [
     path('dataset/<str:dataset_id>/sync_web', views.Dataset.SyncWeb.as_view()),
     path('dataset/<str:dataset_id>/hit_test', views.Dataset.HitTest.as_view()),
     path('dataset/<str:dataset_id>/document', views.Document.as_view(), name='document'),
+    path('dataset/<str:dataset_id>/model', views.Dataset.Model.as_view()),
     path('dataset/document/template/export', views.Template.as_view()),
     path('dataset/document/table_template/export', views.TableTemplate.as_view()),
     path('dataset/<str:dataset_id>/document/web', views.WebDocument.as_view()),
@@ -24,6 +25,7 @@ urlpatterns = [
     path('dataset/<str:dataset_id>/document/batch_hit_handling', views.Document.BatchEditHitHandling.as_view()),
     path('dataset/<str:dataset_id>/document/<int:current_page>/<int:page_size>', views.Document.Page.as_view()),
     path('dataset/<str:dataset_id>/document/batch_refresh', views.Document.BatchRefresh.as_view()),
+    path('dataset/<str:dataset_id>/document/batch_generate_related', views.Document.BatchGenerateRelated.as_view()),
     path('dataset/<str:dataset_id>/document/<str:document_id>', views.Document.Operate.as_view(),
          name="document_operate"),
     path('dataset/document/split', views.Document.Split.as_view(),
@@ -36,12 +38,14 @@ urlpatterns = [
     path('dataset/<str:dataset_id>/document/<str:document_id>/sync', views.Document.SyncWeb.as_view()),
     path('dataset/<str:dataset_id>/document/<str:document_id>/refresh', views.Document.Refresh.as_view()),
     path('dataset/<str:dataset_id>/document/<str:document_id>/paragraph', views.Paragraph.as_view()),
+    path('dataset/<str:dataset_id>/document/batch_generate_related', views.Document.BatchGenerateRelated.as_view()),
     path(
         'dataset/<str:dataset_id>/document/<str:document_id>/paragraph/migrate/dataset/<str:target_dataset_id>/document/<str:target_document_id>',
         views.Paragraph.BatchMigrate.as_view()),
     path('dataset/<str:dataset_id>/document/<str:document_id>/paragraph/_batch', views.Paragraph.Batch.as_view()),
     path('dataset/<str:dataset_id>/document/<str:document_id>/paragraph/<int:current_page>/<int:page_size>',
          views.Paragraph.Page.as_view(), name='paragraph_page'),
+    path('dataset/<str:dataset_id>/document/<str:document_id>/paragraph/batch_generate_related', views.Paragraph.BatchGenerateRelated.as_view()),
     path('dataset/<str:dataset_id>/document/<str:document_id>/paragraph/<paragraph_id>',
          views.Paragraph.Operate.as_view()),
     path('dataset/<str:dataset_id>/document/<str:document_id>/paragraph/<str:paragraph_id>/problem',

apps/dataset/views/dataset.py

@@ -20,6 +20,7 @@ from common.response import result
 from common.response.result import get_page_request_params, get_page_api_response, get_api_response
 from common.swagger_api.common_api import CommonApi
 from dataset.serializers.dataset_serializers import DataSetSerializers
+from setting.serializers.provider_serializers import ModelSerializer
 
 
 class Dataset(APIView):
@@ -223,3 +224,20 @@ class Dataset(APIView):
                       'user_id': str(request.user.id)})
             d.is_valid()
             return result.success(d.page(current_page, page_size))
+
+    class Model(APIView):
+        authentication_classes = [TokenAuth]
+
+        @action(methods=["GET"], detail=False)
+        @has_permissions(ViewPermission(
+            [RoleConstants.ADMIN, RoleConstants.USER],
+            [lambda r, keywords: Permission(group=Group.DATASET, operate=Operate.MANAGE,
+                                            dynamic_tag=keywords.get('dataset_id'))],
+            compare=CompareConstants.AND))
+        def get(self, request: Request, dataset_id: str):
+            print(dataset_id)
+            return result.success(
+                ModelSerializer.Query(
+                    data={'user_id': request.user.id, 'model_type': 'LLM'}).list(
+                    with_valid=True)
+            )

apps/dataset/views/document.py

@@ -393,3 +393,14 @@ class Document(APIView):
                 data={**query_params_to_single_dict(request.query_params), 'dataset_id': dataset_id})
             d.is_valid(raise_exception=True)
             return result.success(d.page(current_page, page_size))
+
+    class BatchGenerateRelated(APIView):
+        authentication_classes = [TokenAuth]
+
+        @action(methods=['PUT'], detail=False)
+        @has_permissions(
+            lambda r, k: Permission(group=Group.DATASET, operate=Operate.MANAGE,
+                                    dynamic_tag=k.get('dataset_id')))
+        def put(self, request: Request, dataset_id: str):
+            return result.success(DocumentSerializers.BatchGenerateRelated(data={'dataset_id': dataset_id})
+                                  .batch_generate_related(request.data))

apps/dataset/views/paragraph.py

@@ -232,3 +232,15 @@ class Paragraph(APIView):
                       'document_id': document_id})
             d.is_valid(raise_exception=True)
             return result.success(d.page(current_page, page_size))
+
+    class BatchGenerateRelated(APIView):
+        authentication_classes = [TokenAuth]
+
+        @action(methods=['PUT'], detail=False)
+        @has_permissions(
+            lambda r, k: Permission(group=Group.DATASET, operate=Operate.MANAGE,
+                                    dynamic_tag=k.get('dataset_id')))
+        def put(self, request: Request, dataset_id: str, document_id: str):
+            return result.success(
+                ParagraphSerializers.BatchGenerateRelated(data={'dataset_id': dataset_id, 'document_id': document_id})
+                .batch_generate_related(request.data))

ui/src/api/dataset.ts

@@ -202,6 +202,22 @@ const exportDataset: (
   return exportExcel(dataset_name + '.xlsx', `dataset/${dataset_id}/export`, undefined, loading)
 }
 
+
+/**
+ * 获取当前用户可使用的模型列表
+ * @param application_id
+ * @param loading
+ * @query  { query_text: string, top_number: number, similarity: number }
+ * @returns
+ */
+const getDatasetModel: (
+  dataset_id: string,
+  loading?: Ref<boolean>
+) => Promise<Result<Array<any>>> = (dataset_id, loading) => {
+  return get(`${prefix}/${dataset_id}/model`, loading)
+}
+
+
 export default {
   getDataset,
   getAllDataset,
@@ -215,5 +231,6 @@ export default {
   putSyncWebDataset,
   putReEmbeddingDataset,
   postQADataset,
-  exportDataset
+  exportDataset,
+  getDatasetModel
 }

ui/src/api/document.ts

@@ -317,6 +317,19 @@ const exportDocument: (
   )
 }
 
+const batchGenerateRelated: (
+  dataset_id: string,
+  data: any,
+  loading?: Ref<boolean>
+) => Promise<Result<boolean>> = (dataset_id, data, loading) => {
+  return put(
+    `${prefix}/${dataset_id}/document/batch_generate_related`,
+    data,
+    undefined,
+    loading
+  )
+}
+
 export default {
   postSplitDocument,
   getDocument,
@@ -338,5 +351,6 @@ export default {
   postQADocument,
   postTableDocument,
   exportDocument,
-  batchRefresh
+  batchRefresh,
+  batchGenerateRelated
 }

ui/src/api/paragraph.ts

@@ -226,6 +226,21 @@ const disassociationProblem: (
   )
 }
 
+const batchGenerateRelated: (
+  dataset_id: string,
+  document_id: string,
+  data: any,
+  loading?: Ref<boolean>
+) => Promise<Result<boolean>> = (dataset_id, document_id, data, loading) => {
+  return put(
+    `${prefix}/${dataset_id}/document/${document_id}/paragraph/batch_generate_related`,
+    data,
+    undefined,
+    loading
+  )
+}
+
+
 export default {
   getParagraph,
   delParagraph,
@@ -236,5 +251,6 @@ export default {
   disassociationProblem,
   associationProblem,
   delMulParagraph,
-  putMigrateMulParagraph
+  putMigrateMulParagraph,
+  batchGenerateRelated
 }

ui/src/views/document/component/GenerateRelatedDialog.vue

@@ -0,0 +1,245 @@
+<template>
+  <el-dialog
+    title="生成关联问题"
+    v-model="dialogVisible"
+    width="600"
+    class="select-dataset-dialog"
+    :close-on-click-modal="false"
+    :close-on-press-escape="false"
+  >
+    <template #header="{ titleId, titleClass }">
+      <div class="my-header flex">
+        <h4 :id="titleId" :class="titleClass">生成关联问题</h4>
+      </div>
+    </template>
+    <div class="content-height">
+      <el-form
+        ref="FormRef"
+        :model="form"
+        :rules="rules"
+        label-position="top"
+        require-asterisk-position="right"
+      >
+        <div class="update-info flex border-r-4 mb-16 w-full">
+          <div class="mt-4">
+            <AppIcon iconName="app-warning-colorful" style="font-size: 16px"></AppIcon>
+          </div>
+          <div class="ml-16 lighter">
+            <p>提示词中的 {data} 为分段内容的占位符，执行时替换为分段内容发送给 AI 模型；</p>
+            <p>AI 模型根据分段内容生成相关问题，请将生成的问题放至&lt;question&gt;&lt;/question&gt;标签中，系统会自动关联标签中的问题；</p>
+            <p>生成效果依赖于所选模型和提示词，用户可自行调整至最佳效果。</p>
+          </div>
+        </div>
+        <el-form-item label="AI 模型" prop="model_id">
+          <el-select
+            v-model="form.model_id"
+            :placeholder="$t('views.application.applicationForm.form.aiModel.placeholder')"
+            class="w-full"
+            popper-class="select-model"
+            :clearable="true"
+          >
+            <el-option-group
+              v-for="(value, label) in modelOptions"
+              :key="value"
+              :label="relatedObject(providerOptions, label, 'provider')?.name"
+            >
+              <el-option
+                v-for="item in value.filter((v: any) => v.status === 'SUCCESS')"
+                :key="item.id"
+                :label="item.name"
+                :value="item.id"
+                class="flex-between"
+              >
+                <div class="flex align-center">
+                        <span
+                          v-html="relatedObject(providerOptions, label, 'provider')?.icon"
+                          class="model-icon mr-8"
+                        ></span>
+                  <span>{{ item.name }}</span>
+                  <el-tag
+                    v-if="item.permission_type === 'PUBLIC'"
+                    type="info"
+                    class="info-tag ml-8"
+                  >公用
+                  </el-tag>
+                </div>
+                <el-icon class="check-icon" v-if="item.id === form.model_id">
+                  <Check />
+                </el-icon>
+              </el-option>
+              <!-- 不可用 -->
+              <el-option
+                v-for="item in value.filter((v: any) => v.status !== 'SUCCESS')"
+                :key="item.id"
+                :label="item.name"
+                :value="item.id"
+                class="flex-between"
+                disabled
+              >
+                <div class="flex">
+                        <span
+                          v-html="relatedObject(providerOptions, label, 'provider')?.icon"
+                          class="model-icon mr-8"
+                        ></span>
+                  <span>{{ item.name }}</span>
+                  <span class="danger">{{
+                      $t('views.application.applicationForm.form.aiModel.unavailable')
+                    }}</span>
+                </div>
+                <el-icon class="check-icon" v-if="item.id === form.model_id">
+                  <Check />
+                </el-icon>
+              </el-option>
+            </el-option-group>
+          </el-select>
+        </el-form-item>
+        <el-form-item label="提示词" prop="prompt">
+          <el-input
+            v-model="form.prompt"
+            placeholder="请输入提示词"
+            :rows="7"
+            type="textarea"
+          />
+        </el-form-item>
+      </el-form>
+
+
+    </div>
+    <template #footer>
+      <span class="dialog-footer">
+        <el-button @click.prevent="dialogVisible = false"> 取消 </el-button>
+        <el-button type="primary" @click="submitHandle(FormRef)" :disabled="!model || loading">
+          确定
+        </el-button>
+      </span>
+    </template>
+  </el-dialog>
+</template>
+<script setup lang="ts">
+import { reactive, ref, watch } from 'vue'
+import { useRoute } from 'vue-router'
+import documentApi from '@/api/document'
+
+import useStore from '@/stores'
+import { relatedObject } from '@/utils/utils'
+import type { Provider } from '@/api/type/model'
+import datasetApi from '@/api/dataset'
+import { groupBy } from 'lodash'
+import { MsgSuccess } from '@/utils/message'
+import { t } from '@/locales'
+import type { FormInstance } from 'element-plus'
+
+const route = useRoute()
+const {
+  params: { id } // id为datasetID
+} = route as any
+
+const { model } = useStore()
+
+
+const emit = defineEmits(['refresh'])
+
+const loading = ref<boolean>(false)
+
+const dialogVisible = ref<boolean>(false)
+const modelOptions = ref<any>(null)
+const providerOptions = ref<Array<Provider>>([])
+const documentIdList = ref<string[]>([])
+
+const FormRef = ref()
+const form = ref({
+  model_id: '',
+  prompt: '内容：{data}\n' +
+    '\n' +
+    '请总结上面的内容，并根据内容总结生成 5 个问题。\n' +
+    '回答要求：\n' +
+    '- 请只输出问题；\n' +
+    '- 请将每个问题放置<question></question>标签中。'
+})
+
+const rules = reactive({
+  model_id: [{ required: true, message: '请选择AI 模型', trigger: 'blur' }],
+  prompt: [{ required: true, message: '请输入提示词', trigger: 'blur' }]
+})
+
+watch(dialogVisible, (bool) => {
+  if (!bool) {
+    form.value.model_id = ''
+  }
+})
+
+const open = (document_ids: string[]) => {
+  getProvider()
+  getModel()
+  documentIdList.value = document_ids
+  dialogVisible.value = true
+}
+
+const submitHandle = async (formEl: FormInstance) => {
+  if (!formEl) {
+    return
+  }
+  await formEl.validate((valid, fields) => {
+    if (valid) {
+      const data = { ...form.value, document_id_list: documentIdList.value }
+      documentApi.batchGenerateRelated(id, data).then(() => {
+        MsgSuccess('生成关联问题成功')
+        emit('refresh')
+        dialogVisible.value = false
+      })
+    }
+  })
+}
+
+
+function getModel() {
+  loading.value = true
+  datasetApi
+    .getDatasetModel(id)
+    .then((res: any) => {
+      modelOptions.value = groupBy(res?.data, 'provider')
+      loading.value = false
+    })
+    .catch(() => {
+      loading.value = false
+    })
+}
+
+function getProvider() {
+  loading.value = true
+  model
+    .asyncGetProvider()
+    .then((res: any) => {
+      providerOptions.value = res?.data
+      loading.value = false
+    })
+    .catch(() => {
+      loading.value = false
+    })
+}
+
+
+defineExpose({ open })
+</script>
+<style lang="scss" scope>
+.select-dataset-dialog {
+  padding: 10px;
+
+  .el-dialog__header {
+    padding: 24px 24px 0 8px;
+  }
+
+  .el-dialog__body {
+    padding: 8px !important;
+  }
+
+  .el-dialog__footer {
+    padding: 0 24px 24px;
+  }
+}
+
+.update-info {
+  background: #d6e2ff;
+  line-height: 25px;
+}
+</style>

ui/src/views/document/index.vue

@@ -26,6 +26,9 @@
             <el-button @click="batchRefresh" :disabled="multipleSelection.length === 0">
               重新向量化
             </el-button>
+            <el-button @click="openGenerateDialog()" :disabled="multipleSelection.length === 0">
+              生成关联问题
+            </el-button>
             <el-button @click="openBatchEditDocument" :disabled="multipleSelection.length === 0">
               设置
             </el-button>
@@ -138,6 +141,9 @@
               <el-text v-else-if="row.status === '3'">
                 <el-icon class="is-loading primary"><Loading /></el-icon> 排队中
               </el-text>
+              <el-text v-else-if="row.status === '4'">
+                <el-icon class="is-loading primary"><Loading /></el-icon> 生成问题中
+              </el-text>
             </template>
           </el-table-column>
           <el-table-column width="130">
@@ -258,6 +264,10 @@
                     </el-button>
                     <template #dropdown>
                       <el-dropdown-menu>
+                        <el-dropdown-item @click="openGenerateDialog(row)">
+                          <el-icon><Connection /></el-icon>
+                          生成关联问题
+                        </el-dropdown-item>
                         <el-dropdown-item @click="openDatasetDialog(row)">
                           <AppIcon iconName="app-migrate"></AppIcon>
                           迁移
@@ -300,6 +310,10 @@
                         <el-dropdown-item icon="Setting" @click="settingDoc(row)"
                           >设置</el-dropdown-item
                         >
+                        <el-dropdown-item @click="openGenerateDialog(row)">
+                          <el-icon><Connection /></el-icon>
+                          生成关联问题
+                        </el-dropdown-item>
                         <el-dropdown-item @click="openDatasetDialog(row)">
                           <AppIcon iconName="app-migrate"></AppIcon>
                           迁移</el-dropdown-item
@@ -324,6 +338,7 @@
       <SyncWebDialog ref="SyncWebDialogRef" @refresh="refresh" />
       <!-- 选择知识库 -->
       <SelectDatasetDialog ref="SelectDatasetDialogRef" @refresh="refreshMigrate" />
+      <GenerateRelatedDialog ref="GenerateRelatedDialogRef" @refresh="refresh" />
     </div>
   </LayoutContainer>
 </template>
@@ -340,6 +355,7 @@ import { datetimeFormat } from '@/utils/time'
 import { hitHandlingMethod } from '@/enums/document'
 import { MsgSuccess, MsgConfirm, MsgError } from '@/utils/message'
 import useStore from '@/stores'
+import GenerateRelatedDialog from '@/views/document/component/GenerateRelatedDialog.vue'
 const router = useRouter()
 const route = useRoute()
 const {
@@ -554,6 +570,19 @@ function batchRefresh() {
   })
 }
 
+function batchGenerateRelated() {
+  const arr: string[] = []
+  multipleSelection.value.map((v) => {
+    if (v) {
+      arr.push(v.id)
+    }
+  })
+  documentApi.batchGenerateRelated(id, arr, loading).then(() => {
+    MsgSuccess('批量生成关联问题成功')
+    multipleTableRef.value?.clearSelection()
+  })
+}
+
 function deleteDocument(row: any) {
   MsgConfirm(
     `是否删除文档：${row.name} ?`,
@@ -643,6 +672,23 @@ function refresh() {
   getList()
 }
 
+const GenerateRelatedDialogRef = ref()
+function openGenerateDialog(row?: any) {
+  const arr: string[] = []
+  if (row) {
+    arr.push(row.id)
+  } else {
+    multipleSelection.value.map((v) => {
+      if (v) {
+        arr.push(v.id)
+      }
+    })
+  }
+
+  GenerateRelatedDialogRef.value.open(arr)
+}
+
+
 onMounted(() => {
   getDetail()
   if (beforePagination.value) {

ui/src/views/paragraph/component/GenerateRelatedDialog.vue

@@ -0,0 +1,246 @@
+<template>
+  <el-dialog
+    title="生成关联问题"
+    v-model="dialogVisible"
+    width="600"
+    class="select-dataset-dialog"
+    :close-on-click-modal="false"
+    :close-on-press-escape="false"
+  >
+    <template #header="{ titleId, titleClass }">
+      <div class="my-header flex">
+        <h4 :id="titleId" :class="titleClass">生成关联问题</h4>
+      </div>
+    </template>
+    <div class="content-height">
+      <el-form
+        ref="FormRef"
+        :model="form"
+        :rules="rules"
+        label-position="top"
+        require-asterisk-position="right"
+      >
+        <div class="update-info flex border-r-4 mb-16 w-full">
+          <div class="mt-4">
+            <AppIcon iconName="app-warning-colorful" style="font-size: 16px"></AppIcon>
+          </div>
+          <div class="ml-16 lighter">
+            <p>提示词中的 {data} 为分段内容的占位符，执行时替换为分段内容发送给 AI 模型；</p>
+            <p>AI
+              模型根据分段内容生成相关问题，请将生成的问题放至&lt;question&gt;&lt;/question&gt;标签中，系统会自动关联标签中的问题；</p>
+            <p>生成效果依赖于所选模型和提示词，用户可自行调整至最佳效果。</p>
+          </div>
+        </div>
+        <el-form-item label="AI 模型" prop="model_id">
+          <el-select
+            v-model="form.model_id"
+            :placeholder="$t('views.application.applicationForm.form.aiModel.placeholder')"
+            class="w-full"
+            popper-class="select-model"
+            :clearable="true"
+          >
+            <el-option-group
+              v-for="(value, label) in modelOptions"
+              :key="value"
+              :label="relatedObject(providerOptions, label, 'provider')?.name"
+            >
+              <el-option
+                v-for="item in value.filter((v: any) => v.status === 'SUCCESS')"
+                :key="item.id"
+                :label="item.name"
+                :value="item.id"
+                class="flex-between"
+              >
+                <div class="flex align-center">
+                        <span
+                          v-html="relatedObject(providerOptions, label, 'provider')?.icon"
+                          class="model-icon mr-8"
+                        ></span>
+                  <span>{{ item.name }}</span>
+                  <el-tag
+                    v-if="item.permission_type === 'PUBLIC'"
+                    type="info"
+                    class="info-tag ml-8"
+                  >公用
+                  </el-tag>
+                </div>
+                <el-icon class="check-icon" v-if="item.id === form.model_id">
+                  <Check />
+                </el-icon>
+              </el-option>
+              <!-- 不可用 -->
+              <el-option
+                v-for="item in value.filter((v: any) => v.status !== 'SUCCESS')"
+                :key="item.id"
+                :label="item.name"
+                :value="item.id"
+                class="flex-between"
+                disabled
+              >
+                <div class="flex">
+                        <span
+                          v-html="relatedObject(providerOptions, label, 'provider')?.icon"
+                          class="model-icon mr-8"
+                        ></span>
+                  <span>{{ item.name }}</span>
+                  <span class="danger">{{
+                      $t('views.application.applicationForm.form.aiModel.unavailable')
+                    }}</span>
+                </div>
+                <el-icon class="check-icon" v-if="item.id === form.model_id">
+                  <Check />
+                </el-icon>
+              </el-option>
+            </el-option-group>
+          </el-select>
+        </el-form-item>
+        <el-form-item label="提示词" prop="prompt">
+          <el-input
+            v-model="form.prompt"
+            placeholder="请输入提示词"
+            :rows="7"
+            type="textarea"
+          />
+        </el-form-item>
+      </el-form>
+
+
+    </div>
+    <template #footer>
+      <span class="dialog-footer">
+        <el-button @click.prevent="dialogVisible = false"> 取消 </el-button>
+        <el-button type="primary" @click="submitHandle(FormRef)" :disabled="!model || loading">
+          确定
+        </el-button>
+      </span>
+    </template>
+  </el-dialog>
+</template>
+<script setup lang="ts">
+import { reactive, ref, watch } from 'vue'
+import { useRoute } from 'vue-router'
+import paragraphApi from '@/api/paragraph'
+
+import useStore from '@/stores'
+import { relatedObject } from '@/utils/utils'
+import type { Provider } from '@/api/type/model'
+import datasetApi from '@/api/dataset'
+import { groupBy } from 'lodash'
+import { MsgSuccess } from '@/utils/message'
+import { t } from '@/locales'
+import type { FormInstance } from 'element-plus'
+
+const route = useRoute()
+const {
+  params: { id, documentId } // id为datasetID
+} = route as any
+
+const { model } = useStore()
+
+
+const emit = defineEmits(['refresh'])
+
+const loading = ref<boolean>(false)
+
+const dialogVisible = ref<boolean>(false)
+const modelOptions = ref<any>(null)
+const providerOptions = ref<Array<Provider>>([])
+const paragraphIdList = ref<string[]>([])
+
+const FormRef = ref()
+const form = ref({
+  model_id: '',
+  prompt: '内容：{data}\n' +
+    '\n' +
+    '请总结上面的内容，并根据内容总结生成 5 个问题。\n' +
+    '回答要求：\n' +
+    '- 请只输出问题；\n' +
+    '- 请将每个问题放置<question></question>标签中。'
+})
+
+const rules = reactive({
+  model_id: [{ required: true, message: '请选择AI 模型', trigger: 'blur' }],
+  prompt: [{ required: true, message: '请输入提示词', trigger: 'blur' }]
+})
+
+watch(dialogVisible, (bool) => {
+  if (!bool) {
+    form.value.model_id = ''
+  }
+})
+
+const open = (paragraph_ids: string[]) => {
+  getProvider()
+  getModel()
+  paragraphIdList.value = paragraph_ids
+  dialogVisible.value = true
+}
+
+const submitHandle = async (formEl: FormInstance) => {
+  if (!formEl) {
+    return
+  }
+  await formEl.validate((valid, fields) => {
+    if (valid) {
+      const data = { ...form.value, paragraph_id_list: paragraphIdList.value }
+      paragraphApi.batchGenerateRelated(id, documentId, data).then(() => {
+        MsgSuccess('生成关联问题成功')
+        emit('refresh')
+        dialogVisible.value = false
+      })
+    }
+  })
+}
+
+
+function getModel() {
+  loading.value = true
+  datasetApi
+    .getDatasetModel(id)
+    .then((res: any) => {
+      modelOptions.value = groupBy(res?.data, 'provider')
+      loading.value = false
+    })
+    .catch(() => {
+      loading.value = false
+    })
+}
+
+function getProvider() {
+  loading.value = true
+  model
+    .asyncGetProvider()
+    .then((res: any) => {
+      providerOptions.value = res?.data
+      loading.value = false
+    })
+    .catch(() => {
+      loading.value = false
+    })
+}
+
+
+defineExpose({ open })
+</script>
+<style lang="scss" scope>
+.select-dataset-dialog {
+  padding: 10px;
+
+  .el-dialog__header {
+    padding: 24px 24px 0 8px;
+  }
+
+  .el-dialog__body {
+    padding: 8px !important;
+  }
+
+  .el-dialog__footer {
+    padding: 0 24px 24px;
+  }
+}
+
+.update-info {
+  background: #d6e2ff;
+  line-height: 25px;
+}
+</style>

ui/src/views/paragraph/index.vue

@@ -121,6 +121,10 @@
                           </el-button>
                           <template #dropdown>
                             <el-dropdown-menu>
+                              <el-dropdown-item @click="openGenerateDialog(item)">
+                                <el-icon><Connection /></el-icon>
+                                生成关联问题</el-dropdown-item
+                              >
                               <el-dropdown-item @click="openSelectDocumentDialog(item)">
                                 <AppIcon iconName="app-migrate"></AppIcon>
                                 迁移</el-dropdown-item
@@ -142,6 +146,9 @@
       </el-scrollbar>
 
       <div class="mul-operation border-t w-full" v-if="isBatch === true">
+        <el-button :disabled="multipleSelection.length === 0" @click="openGenerateDialog()">
+          生成关联问题
+        </el-button>
         <el-button :disabled="multipleSelection.length === 0" @click="openSelectDocumentDialog()">
           迁移
         </el-button>
@@ -154,6 +161,8 @@
     </div>
     <ParagraphDialog ref="ParagraphDialogRef" :title="title" @refresh="refresh" />
     <SelectDocumentDialog ref="SelectDocumentDialogRef" @refresh="refreshMigrateParagraph" />
+    <GenerateRelatedDialog ref="GenerateRelatedDialogRef" @refresh="refresh" />
+
   </LayoutContainer>
 </template>
 <script setup lang="ts">
@@ -166,6 +175,7 @@ import SelectDocumentDialog from './component/SelectDocumentDialog.vue'
 import { numberFormat } from '@/utils/utils'
 import { MsgSuccess, MsgConfirm } from '@/utils/message'
 import useStore from '@/stores'
+import GenerateRelatedDialog from './component/GenerateRelatedDialog.vue'
 const { paragraph } = useStore()
 const route = useRoute()
 const {
@@ -318,6 +328,23 @@ function refresh(data: any) {
   }
 }
 
+
+const GenerateRelatedDialogRef = ref()
+function openGenerateDialog(row?: any) {
+  const arr: string[] = []
+  if (row) {
+    arr.push(row.id)
+  } else {
+    multipleSelection.value.map((v) => {
+      if (v) {
+        arr.push(v)
+      }
+    })
+  }
+
+  GenerateRelatedDialogRef.value.open(arr)
+}
+
 onMounted(() => {
   getDetail()
   getParagraphList()