基于三元组提取领域元数据...

src/main/java/com/supervision/pdfqaserver/domain/ErAttribute.java

@@ -26,11 +26,6 @@ public class ErAttribute implements Serializable {
     /**
      * 属性名
      */
-    private String erName;
-
-    /**
-     * 属性值类型
-     */
     private String attrName;
 
     /**

src/main/java/com/supervision/pdfqaserver/dto/DomainMetadataDTO.java

@@ -2,6 +2,8 @@ package com.supervision.pdfqaserver.dto;
 
 import com.supervision.pdfqaserver.domain.DomainMetadata;
 import lombok.Data;
+
+import java.util.ArrayList;
 import java.util.List;
 
 /**
@@ -28,21 +30,21 @@ public class DomainMetadataDTO {
      */
     private String sourceType;
 
-    private List<ERAttributeDTO> sourceAttributes;
+    private List<ERAttributeDTO> sourceAttributes = new ArrayList<>();
 
     /**
      * 关系
      */
     private String relation;
 
-    private List<ERAttributeDTO> relationAttributes;
+    private List<ERAttributeDTO> relationAttributes = new ArrayList<>();
 
     /**
      * 尾节点类型
      */
     private String targetType;
 
-    private List<ERAttributeDTO> targetAttributes;
+    private List<ERAttributeDTO> targetAttributes = new ArrayList<>();
 
     /**
      * 数据来源：0=手动录入，1=系统自动

src/main/java/com/supervision/pdfqaserver/dto/ERAttributeDTO.java

@@ -1,5 +1,6 @@
 package com.supervision.pdfqaserver.dto;
 
+import com.supervision.pdfqaserver.domain.ErAttribute;
 import lombok.Data;
 
 @Data
@@ -31,4 +32,14 @@ public class ERAttributeDTO {
      * 节点 1 关系 2
      */
     private String erType;
+
+    public ErAttribute toErAttribute() {
+        ErAttribute erAttribute = new ErAttribute();
+        erAttribute.setId(this.id);
+        erAttribute.setDomainMetadataId(this.domainMetadataId);
+        erAttribute.setAttrName(this.attrName);
+        erAttribute.setAttrValueType(this.attrValueType);
+        erAttribute.setErType(this.erType);
+        return erAttribute;
+    }
 }

src/main/java/com/supervision/pdfqaserver/dto/IntentDTO.java

@@ -1,5 +1,6 @@
 package com.supervision.pdfqaserver.dto;
 
+import com.supervision.pdfqaserver.domain.Intention;
 import lombok.Data;
 
 /**
@@ -8,4 +9,34 @@ import lombok.Data;
 @Data
 public class IntentDTO {
 
+    private String id;
+
+    /**
+     * 摘要
+     */
+    private String digest;
+
+    /**
+     * 描述详情
+     */
+    private String desc;
+
+    /**
+     * 领域分类id
+     */
+    private String domainCategoryId;
+
+
+    /**
+     * 数据来源：0=手动录入，1=系统自动
+     */
+    private String generationType;
+
+    public IntentDTO(Intention intention){
+        this.id = intention.getId();
+        this.digest = intention.getDigest();
+        this.desc = intention.getDesc();
+        this.domainCategoryId = intention.getDomainCategoryId();
+        this.generationType = intention.getGenerationType();
+    }
 }

src/main/java/com/supervision/pdfqaserver/mapper/DocumentTruncationMapper.java

@@ -3,6 +3,8 @@ package com.supervision.pdfqaserver.mapper;
 import com.supervision.pdfqaserver.domain.DocumentTruncation;
 import com.baomidou.mybatisplus.core.mapper.BaseMapper;
 
+import java.util.List;
+
 /**
 * @author Administrator
 * @description 针对表【document_truncation(文档切分表)】的数据库操作Mapper
@@ -11,6 +13,7 @@ import com.baomidou.mybatisplus.core.mapper.BaseMapper;
 */
 public interface DocumentTruncationMapper extends BaseMapper<DocumentTruncation> {
 
+    List<DocumentTruncation> listByPdfId(Integer pdfId);
 }
 
 

src/main/java/com/supervision/pdfqaserver/service/DocumentTruncationService.java

@@ -25,4 +25,7 @@ public interface DocumentTruncationService extends IService<DocumentTruncation>
     List<DocumentTruncation> queryByDocumentIds(List<String> documentIds);
 
     List<DocumentTruncation> queryNotERETruncate(List<String> documentIds);
+
+
+    List<DocumentTruncation> listByPdfId(Integer pdfId);
 }

src/main/java/com/supervision/pdfqaserver/service/DomainMetadataService.java

@@ -25,6 +25,12 @@ public interface DomainMetadataService extends IService<DomainMetadata> {
     void saveIfNotExists(DomainMetadata metadata, String domainCategoryId);
 
 
+    /**
+     * 批量保存或更新领域元数据
+     * @param metadatas
+     * @param intentionId
+     * @param domainCategoryId
+     */
     void batchSaveOrUpdateMetadata(List<DomainMetadataDTO> metadatas,String intentionId,String domainCategoryId);
     
     void completeSave(DomainMetadataDTO domainMetadataDTO);

src/main/java/com/supervision/pdfqaserver/service/ErAttributeService.java

@@ -3,6 +3,8 @@ package com.supervision.pdfqaserver.service;
 import com.supervision.pdfqaserver.domain.ErAttribute;
 import com.baomidou.mybatisplus.extension.service.IService;
 
+import java.util.List;
+
 /**
 * @author Administrator
 * @description 针对表【er_attribute(实体关系属性表)】的数据库操作Service
@@ -10,4 +12,9 @@ import com.baomidou.mybatisplus.extension.service.IService;
 */
 public interface ErAttributeService extends IService<ErAttribute> {
 
+
+    void saveIfAbsents(ErAttribute erAttribute, String domainMetadataId);
+
+    List<ErAttribute> listByDomainMetadataId(String domainMetadataId);
+
 }

src/main/java/com/supervision/pdfqaserver/service/IntentionService.java

@@ -31,4 +31,6 @@ public interface IntentionService extends IService<Intention> {
      * @return
      */
     Intention queryByDigestAndDomainCategoryId(String digest, String domainCategoryId);
+
+    List<Intention> queryByDomainCategoryId(String domainCategoryId);
 }

src/main/java/com/supervision/pdfqaserver/service/KnowledgeGraphService.java

@@ -30,6 +30,12 @@ public interface KnowledgeGraphService {
     void generateGraphBaseTrain(Integer pdfId);
 
 
+    /**
+     * 获取三元组转换管道
+     * @param contentType 文档内容类型
+     * @param industry 行业
+     * @return
+     */
     TripleConversionPipeline getTripleConversionPipeline(String contentType,String industry);
 
     void generateGraph(List<EREDTO> eredtoList);

src/main/java/com/supervision/pdfqaserver/service/TripleConversionPipeline.java

@@ -44,7 +44,7 @@ public interface TripleConversionPipeline {
 
 
     /**
-     * 识别出truncate的领域元数据
+     * 识别出truncate的领域元数据, 训练时使用
      * @param truncate 切分文档
      * @return DomainMetadataDTO
      */

src/main/java/com/supervision/pdfqaserver/service/impl/DocumentTruncationServiceImpl.java

@@ -68,6 +68,11 @@ public class DocumentTruncationServiceImpl extends ServiceImpl<DocumentTruncatio
     public List<DocumentTruncation> queryNotERETruncate(List<String> documentIds) {
         return null;
     }
+
+    @Override
+    public List<DocumentTruncation> listByPdfId(Integer pdfId) {
+        return super.baseMapper.listByPdfId(pdfId);
+    }
 }
 
 

src/main/java/com/supervision/pdfqaserver/service/impl/DomainMetadataServiceImpl.java

@@ -5,6 +5,7 @@ import cn.hutool.core.lang.Assert;
 import com.baomidou.mybatisplus.extension.service.impl.ServiceImpl;
 import com.supervision.pdfqaserver.domain.DomainMetadata;
 import com.supervision.pdfqaserver.dto.DomainMetadataDTO;
+import com.supervision.pdfqaserver.dto.ERAttributeDTO;
 import com.supervision.pdfqaserver.service.DomainMetadataService;
 import com.supervision.pdfqaserver.mapper.DomainMetadataMapper;
 import com.supervision.pdfqaserver.service.ErAttributeService;
@@ -31,6 +32,7 @@ public class DomainMetadataServiceImpl extends ServiceImpl<DomainMetadataMapper,
     private final IntentionDomainMetadataService intentionDomainMetadataService;
 
     private final ErAttributeService erAttributeService;
+
     @Override
     public void saveIfNotExists(DomainMetadata metadata) {
 
@@ -71,6 +73,27 @@ public class DomainMetadataServiceImpl extends ServiceImpl<DomainMetadataMapper,
                 this.saveIfNotExists(domainMetadata, domainCategoryId);
                 metadata.setId(domainMetadata.getId());
             }
+            // 保存意图和领域元数据的关系
+            intentionDomainMetadataService.batchSaveIfAbsent(intentionId, List.of(metadata.getId()));
+            // 保存意图和领域元数据的关系属性
+            List<ERAttributeDTO> relationAttributes = metadata.getRelationAttributes();
+            if (CollUtil.isNotEmpty(relationAttributes)){
+                for (ERAttributeDTO relationAttribute : relationAttributes) {
+                    relationAttribute.setDomainMetadataId(metadata.getId());
+                    relationAttribute.setErType("2");
+                    erAttributeService.saveIfAbsents(relationAttribute.toErAttribute(), metadata.getId());
+                }
+            }
+            // 保存意图和领域元数据的节点属性
+            List<ERAttributeDTO> nodeAttributes = metadata.getSourceAttributes();
+            nodeAttributes.addAll(metadata.getTargetAttributes());
+            if (CollUtil.isNotEmpty(nodeAttributes)){
+                for (ERAttributeDTO nodeAttribute : nodeAttributes) {
+                    nodeAttribute.setDomainMetadataId(metadata.getId());
+                    nodeAttribute.setErType("1");
+                    erAttributeService.saveIfAbsents(nodeAttribute.toErAttribute(), metadata.getId());
+                }
+            }
         }
     }
 

src/main/java/com/supervision/pdfqaserver/service/impl/ErAttributeServiceImpl.java

@@ -1,20 +1,44 @@
 package com.supervision.pdfqaserver.service.impl;
 
+import cn.hutool.core.lang.Assert;
+import cn.hutool.core.util.StrUtil;
 import com.baomidou.mybatisplus.extension.service.impl.ServiceImpl;
 import com.supervision.pdfqaserver.domain.ErAttribute;
 import com.supervision.pdfqaserver.service.ErAttributeService;
 import com.supervision.pdfqaserver.mapper.ErAttributeMapper;
+import lombok.extern.slf4j.Slf4j;
 import org.springframework.stereotype.Service;
 
+import java.util.List;
+
 /**
 * @author Administrator
 * @description 针对表【er_attribute(实体关系属性表)】的数据库操作Service实现
 * @createDate 2025-05-14 15:23:54
 */
+@Slf4j
 @Service
 public class ErAttributeServiceImpl extends ServiceImpl<ErAttributeMapper, ErAttribute>
     implements ErAttributeService{
 
+    @Override
+    public void saveIfAbsents(ErAttribute erAttribute, String domainMetadataId) {
+        Assert.notEmpty(domainMetadataId,  "领域分类id不能为空");
+        List<ErAttribute> erAttributes = this.listByDomainMetadataId(domainMetadataId);
+        boolean exists = erAttributes.stream().anyMatch(item -> StrUtil.equals(item.getAttrName(), erAttribute.getAttrName())
+                && StrUtil.equals(item.getAttrValueType(), erAttribute.getAttrValueType()));
+        if (exists){
+            log.info("属性已存在，{},不进行保存...", erAttribute.getAttrName());
+            return;
+        }
+        erAttribute.setDomainMetadataId(domainMetadataId);
+        super.save(erAttribute);
+    }
+
+    @Override
+    public List<ErAttribute> listByDomainMetadataId(String domainMetadataId) {
+        return super.lambdaQuery().eq(ErAttribute::getDomainMetadataId, domainMetadataId).list();
+    }
 }
 
 

src/main/java/com/supervision/pdfqaserver/service/impl/IntentionServiceImpl.java

@@ -60,6 +60,11 @@ public class IntentionServiceImpl extends ServiceImpl<IntentionMapper, Intention
         return this.lambdaQuery().eq(Intention::getDigest, digest)
                 .eq(Intention::getDomainCategoryId, domainCategoryId).one();
     }
+
+    @Override
+    public List<Intention> queryByDomainCategoryId(String domainCategoryId) {
+        return super.lambdaQuery().eq(Intention::getDomainCategoryId, domainCategoryId).list();
+    }
 }
 
 

src/main/java/com/supervision/pdfqaserver/service/impl/KnowledgeGraphServiceImpl.java

@@ -87,12 +87,18 @@ public class KnowledgeGraphServiceImpl implements KnowledgeGraphService {
         PdfInfo pdfInfo = pdfInfoService.getByPdfId(pdfId);
         Assert.notNull(pdfInfo, "pdfId:{}没有找到对应的pdf信息", pdfId);
         if (null == pdfInfo.getTrainStatus()){
+            // todo:训练异常，需要记录异常状态
             log.info("pdfId:{}没有找到对应的pdf训练状态,开始识别文档训练状态...", pdfId);
             pdfInfoService.pdfToGraphStart(pdfId);
             if (StrUtil.isEmpty(pdfInfo.getContentType())){
                 log.info("pdfId:{}没有找到对应的pdf内容类型,开始识别文档内容类型...", pdfId);
                 DocumentContentTypeEnum documentContentTypeEnum = tripleConversionPipeline.makeOutPdfContentType(pdfId);
                 log.info("pdfId:{}识别文档内容类型完成,内容类型:{}", pdfId, documentContentTypeEnum.getType());
+                if (StrUtil.isEmpty(documentContentTypeEnum.getType())){
+                    log.info("pdfId:{}没有找到对应的pdf内容类型,停止后续任务...", pdfId);
+                    pdfInfoService.pdfTrainFail(pdfId);
+                    return;
+                }
                 pdfInfo.setContentType(documentContentTypeEnum.getType());
                 pdfInfoService.updateContentType(pdfId, documentContentTypeEnum.getType());
             }
@@ -100,6 +106,11 @@ public class KnowledgeGraphServiceImpl implements KnowledgeGraphService {
                 log.info("pdfId:{}没有找到对应的pdf行业,开始识别文档行业...", pdfId);
                 String industry = tripleConversionPipeline.makeOutPdfIndustry(pdfId);
                 log.info("pdfId:{}识别文档行业完成,行业:{}", pdfId, industry);
+                if (StrUtil.isEmpty(industry)){
+                    log.info("pdfId:{}没有找到对应的pdf行业,停止后续任务...", pdfId);
+                    pdfInfoService.pdfTrainFail(pdfId);
+                    return;
+                }
                 pdfInfo.setDomainCategoryId(industry);
                 pdfInfoService.updateCategory(pdfId, industry);
             }
@@ -107,43 +118,86 @@ public class KnowledgeGraphServiceImpl implements KnowledgeGraphService {
         TripleConversionPipeline tripleConversionPipeline = this.getTripleConversionPipeline(pdfInfo.getContentType(), pdfInfo.getDomainCategoryId());
 
         List<PdfAnalysisOutput> pdfAnalysisOutputs = pdfAnalysisOutputService.queryByPdfId(pdfId);
+        if (CollUtil.isEmpty(pdfAnalysisOutputs)){
+            log.warn("没有找到pdfId为{}的pdf分析结果,不再进行下一步操作...", pdfId);
+            return;
+        }
         List<String> documentIds = pdfAnalysisOutputs.stream().map(p->String.valueOf(p.getId())).collect(Collectors.toList());
         List<DocumentTruncation> documentTruncations = documentTruncationService.queryByDocumentIds(documentIds);
-        List<TruncateDTO> truncateDTOS = new ArrayList<>();
         if (CollUtil.isNotEmpty(documentTruncations)){
-            log.info("没有找到文档切分数据,pdfId:{},不用重置数据...", pdfId);
-            truncateDTOS = documentTruncations.stream().map(TruncateDTO::new).collect(Collectors.toList());
+            log.info("文档切分数据不为空,pdfId:{},清除切分数据...", pdfId);
+            documentTruncationService.deleteByDocumentIds(documentIds);
         }
-        if (CollUtil.isEmpty(documentTruncations)){
         log.info("开始切割文档切片,pdfId:{}", pdfId);
         List<DocumentDTO> documentDTOList = pdfAnalysisOutputs.stream().map(DocumentDTO::new).collect(Collectors.toList());
-            truncateDTOS = tripleConversionPipeline.sliceDocuments(documentDTOList);
+        List<TruncateDTO> truncateDTOS = tripleConversionPipeline.sliceDocuments(documentDTOList);
         log.info("切割文档切片完成,切片个数:{}", truncateDTOS.size());
         // 保存分片信息
         documentTruncationService.batchSave(truncateDTOS);
-        }
+
         for (TruncateDTO truncateDTO : truncateDTOS) {
+            try {
                 List<String> intents = tripleConversionPipeline.makeOutTruncationIntent(truncateDTO);
                 List<DomainMetadataDTO> domainMetadataDTOS = tripleConversionPipeline.makeOutDomainMetadata(truncateDTO, intents);
                 // 保存意图数据
                 List<Intention> intentions = intentionService.batchSaveIfAbsent(intents, pdfInfo.getDomainCategoryId(), pdfId.toString());
-
                 for (Intention intention : intentions) {
                     List<DomainMetadataDTO> metadataDTOS = domainMetadataDTOS.stream()
                             .filter(d -> StrUtil.equals(d.getIntentDigest(), intention.getDigest())).toList();
                     domainMetadataService.batchSaveOrUpdateMetadata(metadataDTOS,intention.getId(), pdfInfo.getDomainCategoryId());
                 }
-
-
+            }catch (Exception e){
+                log.error("切分文档id:{},意图识别失败", truncateDTO.getId(), e);
             }
 
-
-
+        }
 
     }
 
     @Override
     public void generateGraphBaseTrain(Integer pdfId) {
+        Assert.notNull(pdfId, "pdfId不能为空");
+        PdfInfo pdfInfo = pdfInfoService.getByPdfId(pdfId);
+        Assert.notNull(pdfInfo, "pdfId:{}没有找到对应的pdf信息", pdfId);
+        Assert.isTrue((null !=pdfInfo.getTrainStatus() && pdfInfo.getTrainStatus() == 1),
+                "pdfId:{}的pdf训练状态：{} 不符合要求", pdfId, pdfInfo.getTrainStatus());
+
+        List<TruncateDTO> truncateDTOS = documentTruncationService.listByPdfId(pdfId).stream().map(TruncateDTO::new).collect(Collectors.toList());
+        TripleConversionPipeline conversionPipeline = this.getTripleConversionPipeline(pdfInfo.getContentType(), pdfInfo.getDomainCategoryId());
+        if (CollUtil.isEmpty(truncateDTOS)){
+            log.info("没有找到pdfId为{}的文档切分数据,开始切分数据...", pdfId);
+            List<PdfAnalysisOutput> pdfAnalysisOutputs = pdfAnalysisOutputService.queryByPdfId(pdfId);
+            List<DocumentDTO> documentDTOList = pdfAnalysisOutputs.stream().map(DocumentDTO::new).collect(Collectors.toList());
+            truncateDTOS = conversionPipeline.sliceDocuments(documentDTOList);
+            documentTruncationService.batchSave(truncateDTOS);
+            log.info("切分数据完成,切分个数:{}", truncateDTOS.size());
+        }
+        log.info("开始命名实体识别,切分文档个数:{}", truncateDTOS.size());
+        // 查询当前行业分类下的意图
+        List<IntentDTO> intentionDTOs = intentionService.queryByDomainCategoryId(pdfInfo.getDomainCategoryId()).stream().map(IntentDTO::new).distinct().toList();
+        if (CollUtil.isEmpty(intentionDTOs)){
+            log.info("没有找到行业分类id为{}的意图数据,不再进行下一步操作...", pdfInfo.getDomainCategoryId());
+            return;
+        }
+
+        for (TruncateDTO truncateDTO : truncateDTOS) {
+            try {
+                List<IntentDTO> intents = conversionPipeline.makeOutTruncationIntent(truncateDTO,intentionDTOs);
+                if (CollUtil.isEmpty(intents)){
+                    log.info("切分文档id:{},未正确识别出意图...", truncateDTO.getId());
+                    continue;
+                }
+                EREDTO eredto = conversionPipeline.doEre(truncateDTO, intents);
+                if (null == eredto){
+                    log.info("切分文档id:{},命名实体识别结果为空...", truncateDTO.getId());
+                    continue;
+                }
+                // 保存实体关系抽取结果
+                this.saveERE(eredto, truncateDTO.getId());
+            }catch (Exception e){
+                log.error("命名实体识别失败,切分文档id:{}", truncateDTO.getId(), e);
+            }
+        }
 
     }
 

src/main/resources/mapper/DocumentTruncationMapper.xml

@@ -20,4 +20,10 @@
         layout_type,content,create_time,
         update_time
     </sql>
+    <select id="listByPdfId" resultType="com.supervision.pdfqaserver.domain.DocumentTruncation">
+        select d.*
+        from pdf_analysis_output o
+                 join document_truncation d on o.id::text = d.document_id
+        where o.pdf_id = #{pdfId}
+    </select>
 </mapper>

src/main/resources/mapper/ErAttributeMapper.xml

@@ -7,7 +7,6 @@
     <resultMap id="BaseResultMap" type="com.supervision.pdfqaserver.domain.ErAttribute">
             <id property="id" column="id" jdbcType="VARCHAR"/>
             <result property="domainMetadataId" column="domain_metadata_id" jdbcType="VARCHAR"/>
-            <result property="erName" column="er_name" jdbcType="VARCHAR"/>
             <result property="attrName" column="attr_name" jdbcType="VARCHAR"/>
             <result property="attrValueType" column="attr_value_type" jdbcType="VARCHAR"/>
             <result property="erType" column="er_type" jdbcType="VARCHAR"/>
@@ -16,7 +15,7 @@
     </resultMap>
 
     <sql id="Base_Column_List">
-        id,domain_metadata_id,er_name,
+        id,domain_metadata_id,
         attr_name,attr_value_type,er_type,
         create_time,update_time
     </sql>