fix(ai-web): 完成feedback AI流程

2025-06-16 20:37:14 +08:00
parent e6a1bc5383
commit 45da452f18
7 changed files with 318 additions and 104 deletions
--- a/service-ai/service-ai-web/src/main/java/com/lanyuanxiaoyao/service/ai/web/controller/feedback/FeedbackController.java
+++ b/service-ai/service-ai-web/src/main/java/com/lanyuanxiaoyao/service/ai/web/controller/feedback/FeedbackController.java
@@ -32,7 +32,6 @@ public class FeedbackController {

  @PostMapping("add")
  public void add(@RequestBody CreateItem item) {
-    log.info("Enter method: add[item]. item:{}", item);
    feedbackService.add(item.source, ObjectUtil.defaultIfNull(item.pictures, Lists.immutable.empty()));
  }

@@ -58,6 +57,8 @@ public class FeedbackController {
    private String source;
    private ImmutableList<String> pictures;
    private Feedback.Status status;
+    private String analysis;
+    private String analysisShort;

    public ListItem(FileStoreProperties fileStoreProperties, Feedback feedback) {
      this.id = feedback.getId();
@@ -65,6 +66,8 @@ public class FeedbackController {
      this.pictures = feedback.getPictureIds()
          .collect(id -> StrUtil.format("{}/upload/download/{}", fileStoreProperties.getDownloadPrefix(), id));
      this.status = feedback.getStatus();
+      this.analysis = feedback.getAnalysis();
+      this.analysisShort = feedback.getAnalysisShort();
    }
  }
 }
--- a/service-ai/service-ai-web/src/main/java/com/lanyuanxiaoyao/service/ai/web/entity/Feedback.java
+++ b/service-ai/service-ai-web/src/main/java/com/lanyuanxiaoyao/service/ai/web/entity/Feedback.java
@@ -9,7 +9,7 @@ public class Feedback {
  private String source;
  private String analysisShort;
  private String analysis;
-  private ImmutableList<String> pictureIds;
+  private ImmutableList<Long> pictureIds;
  private Status status;
  private Long createdTime;
  private Long modifiedTime;
--- a/service-ai/service-ai-web/src/main/java/com/lanyuanxiaoyao/service/ai/web/entity/context/FeedbackContext.java
+++ b/service-ai/service-ai-web/src/main/java/com/lanyuanxiaoyao/service/ai/web/entity/context/FeedbackContext.java
@@ -0,0 +1,16 @@
+package com.lanyuanxiaoyao.service.ai.web.entity.context;
+
+import com.lanyuanxiaoyao.service.ai.web.entity.Feedback;
+import java.util.List;
+import lombok.Data;
+import org.eclipse.collections.api.factory.Lists;
+
+/**
+ * @author lanyuanxiaoyao
+ * @version 20250616
+ */
+@Data
+public class FeedbackContext {
+  private Feedback feedback;
+  private List<String> pictureDescriptions = Lists.mutable.empty();
+}
--- a/service-ai/service-ai-web/src/main/java/com/lanyuanxiaoyao/service/ai/web/service/feedback/FeedbackService.java
+++ b/service-ai/service-ai-web/src/main/java/com/lanyuanxiaoyao/service/ai/web/service/feedback/FeedbackService.java
@@ -1,18 +1,24 @@
 package com.lanyuanxiaoyao.service.ai.web.service.feedback;

 import club.kingon.sql.builder.SqlBuilder;
+import cn.hutool.core.lang.Assert;
 import cn.hutool.core.util.EnumUtil;
 import cn.hutool.core.util.ObjectUtil;
 import cn.hutool.core.util.StrUtil;
 import com.lanyuanxiaoyao.service.ai.core.configuration.SnowflakeId;
 import com.lanyuanxiaoyao.service.ai.web.entity.Feedback;
+import com.lanyuanxiaoyao.service.ai.web.entity.context.FeedbackContext;
 import com.lanyuanxiaoyao.service.common.Constants;
+import com.yomahub.liteflow.core.FlowExecutor;
+import java.util.List;
+import java.util.concurrent.TimeUnit;
 import java.util.stream.Collectors;
 import lombok.extern.slf4j.Slf4j;
 import org.eclipse.collections.api.factory.Lists;
 import org.eclipse.collections.api.list.ImmutableList;
 import org.springframework.jdbc.core.JdbcTemplate;
 import org.springframework.jdbc.core.RowMapper;
+import org.springframework.scheduling.annotation.Scheduled;
 import org.springframework.stereotype.Service;
 import org.springframework.transaction.annotation.Transactional;

@@ -20,6 +26,7 @@ import org.springframework.transaction.annotation.Transactional;
@Service
 public class FeedbackService {
  public static final String FEEDBACK_TABLE_NAME = Constants.DATABASE_NAME + ".service_ai_feedback";
+  public static final String[] FEEDBACK_COLUMNS = new String[]{"id", "source", "analysis_short", "analysis", "pictures", "status", "created_time", "modified_time"};
  private static final RowMapper<Feedback> feedbackMapper = (rs, row) -> {
    Feedback feedback = new Feedback();
    feedback.setId(rs.getLong(1));
@@ -29,7 +36,7 @@ public class FeedbackService {
    feedback.setPictureIds(
        StrUtil.isBlank(rs.getString(5))
            ? Lists.immutable.empty()
-            : Lists.immutable.ofAll(StrUtil.split(rs.getString(5), ","))
+            : Lists.immutable.ofAll(StrUtil.split(rs.getString(5), ",")).collect(Long::parseLong)
    );
    feedback.setStatus(EnumUtil.fromString(Feedback.Status.class, rs.getString(6)));
    feedback.setCreatedTime(rs.getTimestamp(7).getTime());
@@ -37,9 +44,38 @@ public class FeedbackService {
    return feedback;
  };
  private final JdbcTemplate template;
+  private final FlowExecutor executor;

-  public FeedbackService(JdbcTemplate template) {
+  @SuppressWarnings("SpringJavaInjectionPointsAutowiringInspection")
+  public FeedbackService(JdbcTemplate template, FlowExecutor executor) {
    this.template = template;
+    this.executor = executor;
+  }
+
+  @Scheduled(initialDelay = 1, fixedDelay = 1, timeUnit = TimeUnit.MINUTES)
+  public void analysis() {
+    List<Feedback> feedbacks = template.query(
+        SqlBuilder.select(FEEDBACK_COLUMNS)
+            .from(FEEDBACK_TABLE_NAME)
+            .whereEq("status", Feedback.Status.ANALYSIS_PROCESSING.name())
+            .build(),
+        feedbackMapper
+    );
+    for (Feedback feedback : feedbacks) {
+      FeedbackContext context = new FeedbackContext();
+      context.setFeedback(feedback);
+      executor.execute2Resp("feedback_analysis", null, context);
+    }
+  }
+
+  public Feedback get(Long id) {
+    return template.queryForObject(
+        SqlBuilder.select(FEEDBACK_COLUMNS)
+            .from(FEEDBACK_TABLE_NAME)
+            .whereEq("id", id)
+            .build(),
+        feedbackMapper
+    );
  }

  @Transactional(rollbackFor = Exception.class)
@@ -55,9 +91,37 @@ public class FeedbackService {
    );
  }

+  @Transactional(rollbackFor = Exception.class)
+  public void updateAnalysis(Long id, String analysis, String analysisShort) {
+    Assert.notNull(id, "ID cannot be null");
+    template.update(
+        SqlBuilder.update(FEEDBACK_TABLE_NAME)
+            .set("analysis", "?")
+            .addSet("analysis_short", "?")
+            .whereEq("id", "?")
+            .precompileSql(),
+        analysis,
+        analysisShort,
+        id
+    );
+  }
+
+  @Transactional(rollbackFor = Exception.class)
+  public void updateStatus(Long id, Feedback.Status status) {
+    Assert.notNull(id, "ID cannot be null");
+    template.update(
+        SqlBuilder.update(FEEDBACK_TABLE_NAME)
+            .set("status", "?")
+            .whereEq("id", "?")
+            .precompileSql(),
+        status.name(),
+        id
+    );
+  }
+
  public ImmutableList<Feedback> list() {
    return template.query(
-            SqlBuilder.select("id", "source", "analysis_short", "analysis", "pictures", "status", "created_time", "modified_time")
+            SqlBuilder.select(FEEDBACK_COLUMNS)
                .from(FEEDBACK_TABLE_NAME)
                .orderByDesc("created_time")
                .build(),
--- a/service-ai/service-ai-web/src/main/java/com/lanyuanxiaoyao/service/ai/web/service/node/FeedbackNodes.java
+++ b/service-ai/service-ai-web/src/main/java/com/lanyuanxiaoyao/service/ai/web/service/node/FeedbackNodes.java
@@ -1,14 +1,181 @@
 package com.lanyuanxiaoyao.service.ai.web.service.node;

+import cn.hutool.core.lang.Assert;
+import cn.hutool.core.util.ObjectUtil;
+import cn.hutool.core.util.StrUtil;
+import com.lanyuanxiaoyao.service.ai.web.entity.Feedback;
+import com.lanyuanxiaoyao.service.ai.web.entity.context.FeedbackContext;
+import com.lanyuanxiaoyao.service.ai.web.entity.vo.DataFileVO;
+import com.lanyuanxiaoyao.service.ai.web.service.DataFileService;
+import com.lanyuanxiaoyao.service.ai.web.service.feedback.FeedbackService;
 import com.yomahub.liteflow.annotation.LiteflowComponent;
+import com.yomahub.liteflow.annotation.LiteflowMethod;
+import com.yomahub.liteflow.core.NodeComponent;
+import com.yomahub.liteflow.enums.LiteFlowMethodEnum;
+import com.yomahub.liteflow.enums.NodeTypeEnum;
+import lombok.extern.slf4j.Slf4j;
 import org.springframework.ai.chat.client.ChatClient;
 import org.springframework.beans.factory.annotation.Qualifier;
+import org.springframework.core.io.FileSystemResource;
+import org.springframework.util.MimeType;
+import org.springframework.util.MimeTypeUtils;

+@Slf4j
@LiteflowComponent
 public class FeedbackNodes {
  private final ChatClient.Builder chatClientBuilder;
+  private final ChatClient.Builder visualChatClientBuilder;
+  private final DataFileService dataFileService;
+  private final FeedbackService feedbackService;

-  public FeedbackNodes(@Qualifier("chat") ChatClient.Builder chatClientBuilder) {
+  public FeedbackNodes(
+      @Qualifier("chat") ChatClient.Builder chatClientBuilder,
+      @Qualifier("visual") ChatClient.Builder visualClientBuilder, DataFileService dataFileService, FeedbackService feedbackService
+  ) {
    this.chatClientBuilder = chatClientBuilder;
+    this.visualChatClientBuilder = visualClientBuilder;
+    this.dataFileService = dataFileService;
+    this.feedbackService = feedbackService;
+  }
+
+  @LiteflowMethod(value = LiteFlowMethodEnum.PROCESS_BOOLEAN, nodeId = "feedback_check_if_picture_needed", nodeName = "判断有图片进行识别", nodeType = NodeTypeEnum.BOOLEAN)
+  public boolean checkIfPictureReadNeeded(NodeComponent node) {
+    FeedbackContext context = node.getContextBean(FeedbackContext.class);
+    Feedback feedback = context.getFeedback();
+    return ObjectUtil.isNotEmpty(feedback.getPictureIds());
+  }
+
+  @LiteflowMethod(value = LiteFlowMethodEnum.PROCESS, nodeId = "image_read", nodeName = "读取图片", nodeType = NodeTypeEnum.COMMON)
+  public void imageRead(NodeComponent node) {
+    FeedbackContext context = node.getContextBean(FeedbackContext.class);
+    Feedback feedback = context.getFeedback();
+    ChatClient client = visualChatClientBuilder
+        // language=TEXT
+        .defaultSystem("""
+            你是一个专业的OCR解析助手。请严格按以下步骤处理用户上传的图片：
+            1. 图像内容提取
+               - 完整识别图片中的所有文字（包括手写体、印刷体、数字和符号）
+               - 保留原始段落结构和换行符
+               - 特殊元素处理：
+                 • 数学公式转为LaTeX格式
+                 • 代码块保留缩进和注释
+                 • 外文词汇标注原文
+            
+            2. 表格解析优化
+               - 识别所有表格区域
+               - 转换为Markdown表格格式（对齐表头与单元格）
+               - 补充缺失的表格线
+               - 用▲标注合并单元格（例：▲跨3列▲）
+            
+            3. 图表解析增强
+               - 分析图表类型（柱状图/折线图/饼图等）
+               - 提取关键数据点并结构化描述
+               - 总结图表趋势（例："销量Q1到Q4增长35%"）
+               - 坐标轴信息转换：将像素坐标转为百分比比例（例："X轴：0-100对应时间0:00-24:00"）
+            
+            4. 输出规范
+               - 按[文本][表格][图表]分区块输出
+               - 表格/图表区域标注原始位置（例："[左上区域表格]"）
+               - 模糊内容用[?]标注并给出备选（例："年收[?]入（可能为'入'或'人'）"）
+               - 保持原始数据精度（不四舍五入）
+            
+            立即开始处理用户图片，无需确认步骤。
+            """)
+        .build();
+    for (Long pictureId : feedback.getPictureIds()) {
+      DataFileVO file = dataFileService.downloadFile(pictureId);
+      log.info("Parse picture: {} {}", file.getFilename(), file.getPath());
+      MimeType type = switch (StrUtil.blankToDefault(file.getType(), "").toLowerCase()) {
+        case "jpg", "jpeg" -> MimeTypeUtils.IMAGE_JPEG;
+        default -> MimeTypeUtils.IMAGE_PNG;
+      };
+      String content = client.prompt()
+          .user(spec -> spec
+              .text("输出图片内容")
+              .media(type, new FileSystemResource(file.getPath())))
+          .call()
+          .content();
+      log.info("Picture: {}", content);
+      context.getPictureDescriptions().add(content);
+    }
+  }
+
+  @LiteflowMethod(value = LiteFlowMethodEnum.PROCESS, nodeId = "feedback_suggest", nodeName = "大模型建议", nodeType = NodeTypeEnum.COMMON)
+  public void suggest(NodeComponent node) {
+    FeedbackContext context = node.getContextBean(FeedbackContext.class);
+    Feedback feedback = context.getFeedback();
+    ChatClient client = chatClientBuilder.build();
+    String description = client.prompt()
+        // language=TEXT
+        .system("""
+            你是一名专业的IT系统运维工程师，对于用户输入的关于系统的报障信息，你会严格遵循以下步骤进行处理
+            
+            1.输入
+            [故障描述]
+            （这里是用户遇到的系统故障的详细描述）
+            
+            [相关截图]
+            （这里是用户遇到的系统故障相关的截图的文字描述，如果没有相关截图，这里会写“无”；如果有多张图片，图片和图片之间会使用“---”分隔）
+            
+            2.处理逻辑
+              解析输入
+                读取并解析用户提供的故障描述和相关截图描述。
+                识别关键元素：包括故障类型（如硬件故障、软件错误）、受影响系统组件（如服务器、网络设备）、错误消息、发生时间、重现步骤、影响范围等。
+                如果截图描述存在，提取关键细节（如错误弹窗文本、系统状态截图），并将其作为辅助证据；如果无截图，则忽略此部分。
+              分析与重写故障描述
+                专业化改写：使用标准IT术语替换非专业用语（例如，“电脑死机”改为“系统无响应”，“连不上网”改为“网络连接中断”），并确保描述符合行业规范。
+                排除歧义：澄清模糊描述（如添加具体时间戳、系统版本、IP地址或错误代码），移除主观语言（如“我觉得”或“可能”），并添加必要上下文（如操作系统环境、相关服务运行状态）。
+                结构化组织：将故障描述重写为逻辑段落，格式包括：
+                  问题概述：简明总结故障本质（例如，“数据库服务异常导致应用无法访问”）。
+                  详细症状：描述具体现象，包括错误消息、发生频率和影响范围（如“影响用户登录功能，错误代码500”）。
+                  重现步骤：列出可复现故障的操作序列（如“1. 访问URL X；2. 触发操作 Y”）。
+                  相关环境：添加系统细节（如“运行在Linux Ubuntu 20.04, Java 11环境”）。
+                整合截图信息：如果截图描述存在，将其嵌入重写中作为证据（例如，“根据截图，错误弹窗显示‘Connection timeout’”）。
+              质量校验
+                检查重写后的内容是否完整、一致且无歧义：确保所有用户输入细节都被涵盖，添加缺失信息（如建议的故障分类），并验证专业术语的准确性。
+                如果输入信息不足（如缺少时间戳或系统版本），在输出中添加注释提示用户补充。
+            
+            3.输出
+              输出一个专业、结构化的故障报告，格式清晰，可直接用于运维团队诊断。
+              重写的故障描述，以结构化段落呈现，涵盖问题概述、详细症状、重现步骤和相关环境。
+              输出将使用中性、客观语言，避免任何个人意见或建议，以确保报告专注于事实描述。
+            """)
+        .call()
+        .content();
+    Assert.notBlank(description, "Description cannot be blank");
+    String analysis = client.prompt()
+        .system("""
+            你是一名专业的IT系统运维工程师，对于用户输入的报障信息，你会给出专业的意见
+            """)
+        .user(StrUtil.format("""
+                [故障描述]
+                {}
+                
+                [相关截图]
+                {}
+                """,
+            description,
+            ObjectUtil.isEmpty(context.getPictureDescriptions()) ? "无" : StrUtil.join(",", context.getPictureDescriptions())
+        ))
+        .call()
+        .content();
+    feedback.setAnalysis(analysis);
+    Assert.notBlank(description, "Analysis cannot be blank");
+    String analysisShort = client.prompt()
+        .system("""
+            你是一名专业的文字编辑，对用户输入的内容，用一段话总结内容
+            """)
+        .user(analysis)
+        .call()
+        .content();
+    feedback.setAnalysisShort(analysisShort);
+  }
+
+  @LiteflowMethod(value = LiteFlowMethodEnum.PROCESS, nodeId = "feedback_save", nodeName = "保存分析内容", nodeType = NodeTypeEnum.COMMON)
+  public void saveFeedback(NodeComponent node) {
+    FeedbackContext context = node.getContextBean(FeedbackContext.class);
+    Feedback feedback = context.getFeedback();
+    feedbackService.updateAnalysis(feedback.getId(), feedback.getAnalysis(), feedback.getAnalysisShort());
+    feedbackService.updateStatus(feedback.getId(), Feedback.Status.ANALYSIS_SUCCESS);
  }
 }
--- a/service-ai/service-ai-web/src/main/resources/liteflow.xml
+++ b/service-ai/service-ai-web/src/main/resources/liteflow.xml
@@ -23,4 +23,14 @@
    <chain id="embedding_submit_directly">
        SER(import_vector_source)
    </chain>
+    <chain id="feedback_analysis">
+        SER(
+            IF(
+                feedback_check_if_picture_needed,
+                image_read
+            ),
+            feedback_suggest,
+            feedback_save
+        )
+    </chain>
 </flow>