feat(aftersales): 售后工单新增问题类型和严重程度字段

云

3 天以前 fb5dcaeb2ab91d0f9ffea26fd15ddcbbe5d36bb9

 yudao-module-ai/src/main/java/cn/iocoder/yudao/module/ai/service/knowledge/AiKnowledgeDocumentServiceImpl.java

@@ -1,226 +1,363 @@
package cn.iocoder.yudao.module.ai.service.knowledge;

import cn.hutool.core.collection.CollUtil;
import cn.hutool.core.util.ObjUtil;
import cn.hutool.core.io.FileUtil;
import cn.hutool.core.io.IoUtil;
import cn.hutool.core.util.ArrayUtil;
import cn.hutool.core.util.StrUtil;
import cn.hutool.http.HttpUtil;
import cn.iocoder.yudao.framework.common.enums.CommonStatusEnum;
import cn.iocoder.yudao.framework.common.pojo.PageResult;
import cn.iocoder.yudao.framework.common.util.object.BeanUtils;
import cn.iocoder.yudao.module.ai.controller.admin.knowledge.vo.document.AiKnowledgeDocumentCreateListReqVO;
import cn.iocoder.yudao.module.ai.controller.admin.knowledge.vo.document.AiKnowledgeDocumentPageReqVO;
import cn.iocoder.yudao.module.ai.controller.admin.knowledge.vo.document.AiKnowledgeDocumentUpdateReqVO;
import cn.iocoder.yudao.module.ai.controller.admin.knowledge.vo.document.AiKnowledgeDocumentUpdateStatusReqVO;
import cn.iocoder.yudao.module.ai.controller.admin.knowledge.vo.knowledge.AiKnowledgeDocumentCreateReqVO;
import cn.iocoder.yudao.module.ai.controller.admin.knowledge.vo.document.*;
import cn.iocoder.yudao.module.ai.controller.admin.knowledge.vo.segment.AiKnowledgeSegmentProcessRespVO;
import cn.iocoder.yudao.module.ai.dal.dataobject.knowledge.AiKnowledgeDO;
import cn.iocoder.yudao.module.ai.dal.dataobject.knowledge.AiKnowledgeDocumentDO;
import cn.iocoder.yudao.module.ai.dal.dataobject.knowledge.AiKnowledgeSegmentDO;
import cn.iocoder.yudao.module.ai.dal.dataobject.model.AiModelDO;
import cn.iocoder.yudao.module.ai.dal.mysql.knowledge.AiKnowledgeDocumentMapper;
import cn.iocoder.yudao.module.ai.dal.mysql.knowledge.AiKnowledgeSegmentMapper;
import cn.iocoder.yudao.module.ai.service.model.AiModelService;
import jakarta.annotation.Resource;
import lombok.extern.slf4j.Slf4j;
import org.springframework.ai.document.Document;
import org.springframework.ai.reader.tika.TikaDocumentReader;
import org.springframework.ai.tokenizer.TokenCountEstimator;
import org.springframework.context.annotation.Lazy;
import org.springframework.core.io.ByteArrayResource;
import org.springframework.ai.transformer.splitter.TokenTextSplitter;
import org.springframework.stereotype.Service;
import org.springframework.transaction.annotation.Transactional;

import org.apache.tika.parser.AutoDetectParser;
import org.apache.tika.parser.ParseContext;
import org.apache.tika.parser.Parser;
import org.apache.tika.extractor.EmbeddedDocumentExtractor;
import org.apache.tika.sax.BodyContentHandler;
import org.xml.sax.ContentHandler;
import java.io.ByteArrayOutputStream;
import java.io.InputStream;
import java.net.HttpURLConnection;
import java.net.URI;
import java.net.URL;
import java.net.URLEncoder;
import java.nio.charset.StandardCharsets;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.List;

import static cn.iocoder.yudao.framework.common.exception.util.ServiceExceptionUtil.exception;
import static cn.iocoder.yudao.framework.common.util.collection.CollectionUtils.convertList;
import static cn.iocoder.yudao.module.ai.enums.ErrorCodeConstants.*;

/**
 * AI 知识库文档 Service 实现类
 *
 * @author xiaoxin
 */
@Service
@Slf4j
@Service
public class AiKnowledgeDocumentServiceImpl implements AiKnowledgeDocumentService {

    @Resource
    private AiKnowledgeDocumentMapper knowledgeDocumentMapper;

    private AiKnowledgeDocumentMapper documentMapper;
    @Resource
    private TokenCountEstimator tokenCountEstimator;

    private AiKnowledgeSegmentMapper segmentMapper;
    @Resource
    private AiKnowledgeSegmentService knowledgeSegmentService;
    @Resource
    @Lazy // 延迟加载，避免循环依赖
    private AiKnowledgeService knowledgeService;
    @Resource
    private AiKnowledgeSegmentService segmentService;
    @Resource
    private AiModelService modelService;

    @Override
    public Long createKnowledgeDocument(AiKnowledgeDocumentCreateReqVO createReqVO) {
        // 1. 校验参数
        knowledgeService.validateKnowledgeExists(createReqVO.getKnowledgeId());
    @Transactional(rollbackFor = Exception.class)
    public List<Long> createDocuments(AiKnowledgeDocumentCreateReqVO createReqVO) {
        AiKnowledgeDO knowledge = knowledgeService.validateKnowledge(createReqVO.getKnowledgeId());
        List<AiKnowledgeDocumentCreateReqVO.DocumentItem> list = createReqVO.getList();
        if (CollUtil.isEmpty(list)) {
            throw new IllegalArgumentException("文档列表不能为空");
        }
        int defaultSegmentMaxTokens = createReqVO.getSegmentMaxTokens() != null
                ? createReqVO.getSegmentMaxTokens() : 800;

        // 2. 下载文档
        String content = readUrl(createReqVO.getUrl());
        List<Long> ids = new ArrayList<>();
        for (AiKnowledgeDocumentCreateReqVO.DocumentItem item : list) {
            AiKnowledgeDocumentCreateReqVO.UrlInfo urlInfo = item.getUrl();
            String fileUrl = urlInfo != null ? urlInfo.getUrl() : null;
            AiKnowledgeDocumentDO document = new AiKnowledgeDocumentDO();
            document.setKnowledgeId(knowledge.getId());
            document.setName(item.getName());
            document.setUrl(fileUrl);
            document.setStatus(CommonStatusEnum.ENABLE.getStatus());
            document.setSegmentMaxTokens(defaultSegmentMaxTokens);
            documentMapper.insert(document);
            ids.add(document.getId());

        // 3. 文档记录入库
        AiKnowledgeDocumentDO documentDO = BeanUtils.toBean(createReqVO, AiKnowledgeDocumentDO.class)
                .setContent(content).setContentLength(content.length()).setTokens(tokenCountEstimator.estimate(content))
                .setStatus(CommonStatusEnum.ENABLE.getStatus());
        knowledgeDocumentMapper.insert(documentDO);

        // 4. 文档切片入库（异步）
        knowledgeSegmentService.createKnowledgeSegmentBySplitContentAsync(documentDO.getId(), content);
        return documentDO.getId();
            if (fileUrl != null) {
                try {
                    loadDocumentContent(document);
                    documentMapper.updateById(document);
                    // 自动执行分段和向量化
                    processDocumentSegmentsInternal(document, knowledge);
                } catch (Exception e) {
                    log.error("文档[{}]处理失败: {}", document.getId(), e.getMessage());
                    document.setStatus(CommonStatusEnum.DISABLE.getStatus());
                    documentMapper.updateById(document);
                }
            }
        }
        return ids;
    }

    @Override
    public List<Long> createKnowledgeDocumentList(AiKnowledgeDocumentCreateListReqVO createListReqVO) {
        // 1. 校验参数
        knowledgeService.validateKnowledgeExists(createListReqVO.getKnowledgeId());

        // 2. 下载文档
        List<String> contents = convertList(createListReqVO.getList(), document -> readUrl(document.getUrl()));

        // 3. 文档记录入库
        List<AiKnowledgeDocumentDO> documentDOs = new ArrayList<>(createListReqVO.getList().size());
        for (int i = 0; i < createListReqVO.getList().size(); i++) {
            AiKnowledgeDocumentCreateListReqVO.Document documentVO = createListReqVO.getList().get(i);
            String content = contents.get(i);
            documentDOs.add(BeanUtils.toBean(documentVO, AiKnowledgeDocumentDO.class)
                    .setKnowledgeId(createListReqVO.getKnowledgeId())
                    .setContent(content).setContentLength(content.length())
                    .setTokens(tokenCountEstimator.estimate(content))
                    .setSegmentMaxTokens(createListReqVO.getSegmentMaxTokens())
                    .setStatus(CommonStatusEnum.ENABLE.getStatus()));
    private void loadDocumentContent(AiKnowledgeDocumentDO document) {
        try {
            byte[] fileBytes = downloadFile(document.getUrl());
            String content = extractText(fileBytes, document.getName());
            if (StrUtil.isEmpty(content)) {
                content = new String(fileBytes, java.nio.charset.Charset.forName("UTF-8"));
        }
        knowledgeDocumentMapper.insertBatch(documentDOs);

        // 4. 批量创建文档切片（异步）
        documentDOs.forEach(documentDO -> knowledgeSegmentService
                .createKnowledgeSegmentBySplitContentAsync(documentDO.getId(), documentDO.getContent()));
        return convertList(documentDOs, AiKnowledgeDocumentDO::getId);
    }

    @Override
    public PageResult<AiKnowledgeDocumentDO> getKnowledgeDocumentPage(AiKnowledgeDocumentPageReqVO pageReqVO) {
        return knowledgeDocumentMapper.selectPage(pageReqVO);
    }

    @Override
    public AiKnowledgeDocumentDO getKnowledgeDocument(Long id) {
        return knowledgeDocumentMapper.selectById(id);
    }

    @Override
    public void updateKnowledgeDocument(AiKnowledgeDocumentUpdateReqVO reqVO) {
        // 1. 校验文档是否存在
        AiKnowledgeDocumentDO oldDocument = validateKnowledgeDocumentExists(reqVO.getId());

        // 2. 更新文档
        AiKnowledgeDocumentDO document = BeanUtils.toBean(reqVO, AiKnowledgeDocumentDO.class);
        knowledgeDocumentMapper.updateById(document);

        // 3. 如果处于开启状态，并且最大 tokens 发生变化，则 segment 需要重新索引
        if (CommonStatusEnum.isEnable(oldDocument.getStatus())
                && reqVO.getSegmentMaxTokens() != null
                && ObjUtil.notEqual(reqVO.getSegmentMaxTokens(), oldDocument.getSegmentMaxTokens())) {
            // 删除旧的文档切片
            knowledgeSegmentService.deleteKnowledgeSegmentByDocumentId(reqVO.getId());
            // 重新创建文档切片
            knowledgeSegmentService.createKnowledgeSegmentBySplitContentAsync(reqVO.getId(), oldDocument.getContent());
            document.setContent(content);
            document.setContentLength(content.length());
            document.setTokens(estimateTokens(content));
        } catch (Exception e) {
            throw new RuntimeException("文档内容加载失败: " + e.getMessage(), e);
        }
    }

    @Override
    public void updateKnowledgeDocumentStatus(AiKnowledgeDocumentUpdateStatusReqVO reqVO) {
        // 1. 校验存在
        AiKnowledgeDocumentDO document = validateKnowledgeDocumentExists(reqVO.getId());
    private byte[] downloadFile(String fileUrl) {
        try {
            // 手动编码 URL 路径中的非 ASCII 字符，避免 URI 构造函数报错
            String encodedUrl = encodeUrlPath(fileUrl);
            URL url = URL.of(new URI(encodedUrl), null);
            HttpURLConnection conn = (HttpURLConnection) url.openConnection();
            conn.setConnectTimeout(30000);
            conn.setReadTimeout(60000);
            conn.setRequestMethod("GET");
            try (InputStream is = conn.getInputStream();
                 ByteArrayOutputStream bos = new ByteArrayOutputStream()) {
                IoUtil.copy(is, bos);
                return bos.toByteArray();
            }
        } catch (Exception e) {
            throw new RuntimeException("文件下载失败: " + e.getMessage(), e);
        }
    }

        // 2. 更新状态
        knowledgeDocumentMapper.updateById(new AiKnowledgeDocumentDO()
                .setId(reqVO.getId()).setStatus(reqVO.getStatus()));
    private static String encodeUrlPath(String urlString) {
        // 分离 scheme://authority 和 path?query#fragment
        int schemeEnd = urlString.indexOf("://");
        if (schemeEnd < 0) return urlString;
        int pathStart = urlString.indexOf('/', schemeEnd + 3);
        if (pathStart < 0) return urlString; // 无 path

        // 3. 处理文档切片
        if (CommonStatusEnum.isEnable(reqVO.getStatus())) {
            knowledgeSegmentService.createKnowledgeSegmentBySplitContentAsync(reqVO.getId(), document.getContent());
        String base = urlString.substring(0, pathStart);
        String pathAndRest = urlString.substring(pathStart);

        // 分离 path 和 query + fragment
        int queryStart = pathAndRest.indexOf('?');
        int fragStart = pathAndRest.indexOf('#');
        String rawPath, query, fragment;
        if (queryStart >= 0) {
            rawPath = pathAndRest.substring(0, queryStart);
            if (fragStart >= 0 && fragStart > queryStart) {
                query = pathAndRest.substring(queryStart, fragStart);
                fragment = pathAndRest.substring(fragStart);
        } else {
            knowledgeSegmentService.deleteKnowledgeSegmentByDocumentId(reqVO.getId());
                query = pathAndRest.substring(queryStart);
                fragment = "";
            }
        } else if (fragStart >= 0) {
            rawPath = pathAndRest.substring(0, fragStart);
            query = "";
            fragment = pathAndRest.substring(fragStart);
        } else {
            rawPath = pathAndRest;
            query = "";
            fragment = "";
        }

        // 对路径每个段做 URL 编码
        StringBuilder encodedPath = new StringBuilder();
        for (String segment : rawPath.split("/")) {
            if (!segment.isEmpty()) {
                encodedPath.append("/").append(URLEncoder.encode(segment, StandardCharsets.UTF_8)
                        .replace("+", "%20"));
            }
        }
        if (rawPath.endsWith("/")) encodedPath.append("/");

        return base + encodedPath + query + fragment;
    }

    private String extractText(byte[] fileBytes, String fileName) {
        try {
            String ext = FileUtil.extName(fileName).toLowerCase();
            if (ArrayUtil.contains(new String[]{"txt", "md", "json", "xml", "csv", "yaml", "yml"}, ext)) {
                return new String(fileBytes, StandardCharsets.UTF_8);
            }
            // 使用 AutoDetectParser + EmbeddedDocumentExtractor 跳过嵌入图片，避免提取到二进制乱码
            Parser parser = new AutoDetectParser();
            ParseContext context = new ParseContext();
            context.set(EmbeddedDocumentExtractor.class, new EmbeddedDocumentExtractor() {
                @Override
                public boolean shouldParseEmbedded(org.apache.tika.metadata.Metadata metadata) {
                    return false;
                }
                @Override
                public void parseEmbedded(InputStream inputStream, ContentHandler contentHandler,
                                          org.apache.tika.metadata.Metadata metadata, boolean outputHtml) {
                }
            });
            BodyContentHandler handler = new BodyContentHandler(-1);
            try (InputStream is = new java.io.ByteArrayInputStream(fileBytes)) {
                parser.parse(is, handler, new org.apache.tika.metadata.Metadata(), context);
            }
            return handler.toString().trim();
        } catch (Exception e) {
            log.warn("Tika 解析文档失败，尝试 UTF-8 文本读取: {}", e.getMessage());
            return "";
        }
    }

    @Override
    @Transactional(rollbackFor = Exception.class)
    public void deleteKnowledgeDocument(Long id) {
        // 1. 校验存在
        validateKnowledgeDocumentExists(id);

        // 2. 删除
        knowledgeDocumentMapper.deleteById(id);

        // 3. 删除对应的段落
        knowledgeSegmentService.deleteKnowledgeSegmentByDocumentId(id);
    }

    @Override
    public AiKnowledgeDocumentDO validateKnowledgeDocumentExists(Long id) {
        AiKnowledgeDocumentDO knowledgeDocument = knowledgeDocumentMapper.selectById(id);
        if (knowledgeDocument == null) {
            throw exception(KNOWLEDGE_DOCUMENT_NOT_EXISTS);
        }
        return knowledgeDocument;
    }

    @Override
    public String readUrl(String url) {
        // 下载文件
        ByteArrayResource resource;
    public void updateDocument(AiKnowledgeDocumentUpdateReqVO updateReqVO) {
        AiKnowledgeDocumentDO document = validateDocumentExists(updateReqVO.getId());
        boolean urlChanged = StrUtil.isNotEmpty(updateReqVO.getUrl())
                && !updateReqVO.getUrl().equals(document.getUrl());
        if (StrUtil.isNotEmpty(updateReqVO.getName())) document.setName(updateReqVO.getName());
        if (urlChanged) document.setUrl(updateReqVO.getUrl());
        documentMapper.updateById(document);
        // URL 变更时重新提取内容、重新分段和向量化
        if (urlChanged) {
        try {
            byte[] bytes = HttpUtil.downloadBytes(url);
            if (bytes.length == 0) {
                loadDocumentContent(document);
                documentMapper.updateById(document);
                AiKnowledgeDO knowledge = knowledgeService.validateKnowledge(document.getKnowledgeId());
                // 删除旧分段和向量
                segmentService.deleteSegmentsByDocumentId(document.getId());
                // 重新分段 + 向量化
                processDocumentSegmentsInternal(document, knowledge);
            } catch (Exception e) {
                log.error("文档[{}] URL 更新后处理失败: {}", document.getId(), e.getMessage());
                document.setStatus(CommonStatusEnum.DISABLE.getStatus());
                documentMapper.updateById(document);
            }
        }
    }

    @Override
    @Transactional(rollbackFor = Exception.class)
    public void deleteDocument(Long id) {
        AiKnowledgeDocumentDO document = validateDocumentExists(id);
        segmentService.deleteSegmentsByDocumentId(id);
        documentMapper.deleteById(id);
    }

    @Override
    public AiKnowledgeDocumentDO getDocument(Long id) {
        return documentMapper.selectById(id);
    }

    @Override
    public AiKnowledgeDocumentDO validateDocumentExists(Long id) {
        AiKnowledgeDocumentDO document = documentMapper.selectById(id);
        if (document == null) throw exception(KNOWLEDGE_DOCUMENT_NOT_EXISTS);
        return document;
    }

    @Override
    public PageResult<AiKnowledgeDocumentDO> getDocumentPage(AiKnowledgeDocumentPageReqVO pageReqVO) {
        return documentMapper.selectPage(pageReqVO);
    }

    @Override
    @Transactional(rollbackFor = Exception.class)
    public void updateDocumentStatus(AiKnowledgeDocumentUpdateStatusReqVO updateStatusReqVO) {
        AiKnowledgeDocumentDO document = validateDocumentExists(updateStatusReqVO.getId());
        document.setStatus(updateStatusReqVO.getStatus());
        documentMapper.updateById(document);
    }

    @Override
    public List<AiKnowledgeSegmentProcessRespVO> getDocumentProcessingProgress(List<Long> documentIds) {
        if (CollUtil.isEmpty(documentIds)) return Collections.emptyList();
        return segmentMapper.selectProcessList(documentIds);
    }

    @Override
    public void processDocumentSegments(Long documentId) {
        AiKnowledgeDocumentDO document = validateDocumentExists(documentId);
        if (StrUtil.isEmpty(document.getContent())) {
                throw exception(KNOWLEDGE_DOCUMENT_FILE_EMPTY);
            }
            resource = new ByteArrayResource(bytes);
        } catch (Exception e) {
            log.error("[readUrl][url({}) 读取失败]", url, e);
            throw exception(KNOWLEDGE_DOCUMENT_FILE_DOWNLOAD_FAIL);
        AiKnowledgeDO knowledge = knowledgeService.validateKnowledge(document.getKnowledgeId());
        processDocumentSegmentsInternal(document, knowledge);
        }

        // 读取文件
        TikaDocumentReader loader = new TikaDocumentReader(resource);
        List<Document> documents = loader.get();
        Document document = CollUtil.getFirst(documents);
        if (document == null || StrUtil.isEmpty(document.getText())) {
            throw exception(KNOWLEDGE_DOCUMENT_FILE_READ_FAIL);
    private void processDocumentSegmentsInternal(AiKnowledgeDocumentDO document, AiKnowledgeDO knowledge) {
        if (StrUtil.isEmpty(document.getContent())) return;

        AiModelDO embeddingModel = modelService.validateModel(knowledge.getEmbeddingModelId());

        // 删除旧分段
        segmentService.deleteSegmentsByDocumentId(document.getId());

        // 文本切片
        int segmentMaxTokens = document.getSegmentMaxTokens() != null ? document.getSegmentMaxTokens() : 800;
        List<String> segmentTexts = splitContent(document.getContent(), segmentMaxTokens);
        if (CollUtil.isEmpty(segmentTexts)) return;

        // 向量化并存储
        List<AiKnowledgeSegmentDO> segments = new ArrayList<>();
        for (String content : segmentTexts) {
            if (StrUtil.isEmpty(content.trim())) continue;
            segments.add(buildSegmentDO(knowledge.getId(), document.getId(), content));
        }
        return document.getText();
        segmentService.saveSegments(segments, knowledge.getId(), embeddingModel.getId());
    }

    private List<String> splitContent(String content, int maxTokens) {
        TokenTextSplitter splitter = TokenTextSplitter.builder()
                .withChunkSize(maxTokens)
                .withMinChunkSizeChars(50)
                .withMinChunkLengthToEmbed(10)
                .withMaxNumChunks(1000)
                .withKeepSeparator(true)
                .build();
        List<Document> docs = splitter.apply(Collections.singletonList(new Document(content)));
        List<String> result = new ArrayList<>();
        for (Document doc : docs) {
            if (StrUtil.isNotEmpty(doc.getText())) {
                result.add(doc.getText().trim());
            }
        }
        return result;
    }

    @Override
    public List<AiKnowledgeDocumentDO> getKnowledgeDocumentList(Collection<Long> ids) {
        if (CollUtil.isEmpty(ids)) {
            return Collections.emptyList();
    public List<String> previewSplit(String url, String name, Integer segmentMaxTokens) {
        byte[] fileBytes = downloadFile(url);
        String text = extractText(fileBytes, name);
        if (StrUtil.isEmpty(text)) {
            text = new String(fileBytes, java.nio.charset.Charset.forName("UTF-8"));
        }
        return knowledgeDocumentMapper.selectByIds(ids);
        int maxTokens = segmentMaxTokens != null && segmentMaxTokens > 0 ? segmentMaxTokens : 800;
        return splitContent(text, maxTokens);
    }

    @Override
    public List<AiKnowledgeDocumentDO> getKnowledgeDocumentListByKnowledgeId(Long knowledgeId) {
        return knowledgeDocumentMapper.selectListByKnowledgeId(knowledgeId);
    private AiKnowledgeSegmentDO buildSegmentDO(Long knowledgeId, Long documentId, String content) {
        AiKnowledgeSegmentDO segment = new AiKnowledgeSegmentDO();
        segment.setKnowledgeId(knowledgeId);
        segment.setDocumentId(documentId);
        segment.setContent(content);
        segment.setContentLength(content.length());
        segment.setTokens(estimateTokens(content));
        segment.setStatus(CommonStatusEnum.ENABLE.getStatus());
        segment.setVectorId(AiKnowledgeSegmentDO.VECTOR_ID_EMPTY);
        segment.setRetrievalCount(0);
        return segment;
    }

    @Override
    @Transactional(rollbackFor = Exception.class)
    public void deleteKnowledgeDocumentByKnowledgeId(Long knowledgeId) {
        // 1. 获取该知识库下的所有文档
        List<AiKnowledgeDocumentDO> documents = knowledgeDocumentMapper.selectListByKnowledgeId(knowledgeId);
        if (CollUtil.isEmpty(documents)) {
            return;
    private Integer estimateTokens(String text) {
        if (StrUtil.isEmpty(text)) return 0;
        int chineseChars = 0, englishWords = 0;
        for (char c : text.toCharArray()) {
            if (c >= 0x4E00 && c <= 0x9FA5) chineseChars++;
        }

        // 2. 逐个删除文档及其对应的段落
        for (AiKnowledgeDocumentDO document : documents) {
            deleteKnowledgeDocument(document.getId());
        for (String word : text.split("\\s+")) {
            if (word.matches(".*[a-zA-Z].*")) englishWords++;
        }
        return chineseChars + (int) (englishWords * 1.3);
    }

}