当前位置：首页 > news >正文

511-Spring AI Alibaba Graph 智能写作助手示例

news 2025/11/2 9:39:41

本示例展示如何使用 Spring AI Alibaba Graph 构建一个智能写作助手工作流，实现文本摘要生成、内容改写和标题生成的自动化流程。

1. 示例目标

我们将创建一个基于有向图工作流的智能写作助手，包含以下核心功能：

文本摘要生成：对输入的原始文本自动生成简洁摘要。
摘要反馈判断：通过大模型判断摘要质量，决定是否需要重新生成。
内容改写优化：对通过的摘要进行语言优化和润色。
标题自动生成：基于优化后的内容生成吸引人的标题。
工作流可视化：自动生成 PlantUML 流程图，便于理解整个流程。

2. 技术栈与核心依赖

Spring Boot 3.x
Spring AI Alibaba Graph (用于构建有向图工作流)
Spring AI Alibaba DashScope (用于对接阿里云通义大模型)
Maven (项目构建工具)

在 pom.xml 中，你需要引入以下核心依赖：

<dependencies><!-- Spring Web 用于构建 RESTful API --><dependency><groupId>org.springframework.boot</groupId><artifactId>spring-boot-starter-web</artifactId></dependency><!-- Spring AI Alibaba DashScope 启动器 --><dependency><groupId>com.alibaba.cloud.ai</groupId><artifactId>spring-ai-alibaba-starter-dashscope</artifactId><version>${spring-ai-alibaba.version}</version></dependency><!-- Spring AI Alibaba Graph 核心库 --><dependency><groupId>com.alibaba.cloud.ai</groupId><artifactId>spring-ai-alibaba-graph-core</artifactId><version>1.0.0.3</version></dependency>
</dependencies>

3. 项目配置

在 src/main/resources/application.yml 文件中，配置你的 DashScope API Key。

server:port: 8080spring:application:name: spring-ai-alibaba-graph-exampleai:dashscope:api-key: ${AI_DASHSCOPE_API_KEY}chat:options:model: qwen-plus

重要提示：请将 AI_DASHSCOPE_API_KEY 环境变量设置为你从阿里云获取的有效 API Key。

4. 工作流设计

智能写作助手的工作流由以下几个节点组成：

summarizer（摘要生成器）：接收原始文本，生成简洁摘要。
feedback_classifier（反馈分类器）：评估摘要质量，返回 "positive" 或 "negative"。
reworder（内容改写器）：对通过的摘要进行语言优化。
title_generator（标题生成器）：基于优化后的内容生成标题。

4.1 工作流流程图

以下是该工作流的 PlantUML 流程图：

@startuml
skinparam usecaseFontSize 14
skinparam usecaseStereotypeFontSize 12
skinparam hexagonFontSize 14
skinparam hexagonStereotypeFontSize 12
title "writing assistant flow"
footer
powered by spring-ai-alibaba
end footer
circle start<<input>> as __START__
circle stop as __END__
usecase "summarizer"<<Node>>
usecase "feedback_classifier"<<Node>>
usecase "reworder"<<Node>>
usecase "title_generator"<<Node>>
hexagon "check state" as condition1<<Condition>>
"__START__" -down-> "summarizer"
"summarizer" -down-> "feedback_classifier"
"feedback_classifier" .down.> "condition1"
"condition1" .down.> "summarizer": "negative"
"condition1" .down.> "reworder": "positive"
"reworder" -down-> "title_generator"
"title_generator" -down-> "__END__"
@enduml

5. 核心代码实现

5.1 主应用类

Application.java 是标准的 Spring Boot 启动类：

package com.alibaba.example;import org.springframework.boot.SpringApplication;
import org.springframework.boot.autoconfigure.SpringBootApplication;@SpringBootApplication
public class Application {public static void main(String[] args) {SpringApplication.run(Application.class, args);}
}

5.2 控制器实现

WritingAssistantController.java 提供了 REST API 接口：

package com.alibaba.example.controller;import com.alibaba.cloud.ai.graph.CompiledGraph;
import com.alibaba.cloud.ai.graph.StateGraph;
import com.alibaba.cloud.ai.graph.exception.GraphRunnerException;
import com.alibaba.cloud.ai.graph.exception.GraphStateException;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.beans.factory.annotation.Qualifier;
import org.springframework.web.bind.annotation.*;import java.util.Map;@RestController
@RequestMapping("/write")
public class WritingAssistantController {private final CompiledGraph compiledGraph;@Autowiredpublic WritingAssistantController(@Qualifier("writingAssistantGraph") StateGraph writingAssistantGraph)throws GraphStateException {this.compiledGraph = writingAssistantGraph.compile();}/*** 调用写作助手流程图 示例请求：GET /write?text=今天我去了西湖，天气特别好，感觉特别开心*/@GetMappingpublic Map<String, Object> write(@RequestParam("text") String inputText) throws GraphRunnerException {var resultFuture = compiledGraph.invoke(Map.of("original_text", inputText));var result = resultFuture.get();return result.data();}
}

5.3 工作流配置

WritingAssistantAutoconfiguration.java 是整个示例的核心，定义了工作流的结构和各个节点的行为：

@Configuration
public class WritingAssistantAutoconfiguration {@Beanpublic StateGraph writingAssistantGraph(ChatModel chatModel) throws GraphStateException {ChatClient chatClient = ChatClient.builder(chatModel).defaultAdvisors(new SimpleLoggerAdvisor()).build();KeyStrategyFactory keyStrategyFactory = () -> {HashMap<String, KeyStrategy> keyStrategyHashMap = new HashMap<>();keyStrategyHashMap.put("original_text", new ReplaceStrategy());keyStrategyHashMap.put("summary", new ReplaceStrategy());keyStrategyHashMap.put("summary_feedback", new ReplaceStrategy());keyStrategyHashMap.put("reworded", new ReplaceStrategy());keyStrategyHashMap.put("title", new ReplaceStrategy());return keyStrategyHashMap;};StateGraph graph = new StateGraph(keyStrategyFactory).addNode("summarizer", node_async(new SummarizerNode(chatClient))).addNode("feedback_classifier", node_async(new SummaryFeedbackClassifierNode(chatClient, "summary"))).addNode("reworder", node_async(new RewordingNode(chatClient))).addNode("title_generator", node_async(new TitleGeneratorNode(chatClient))).addEdge(START, "summarizer").addEdge("summarizer", "feedback_classifier").addConditionalEdges("feedback_classifier", edge_async(new FeedbackDispatcher()),Map.of("positive", "reworder", "negative", "summarizer")).addEdge("reworder", "title_generator").addEdge("title_generator", END);// 添加 PlantUML 打印GraphRepresentation representation = graph.getGraph(GraphRepresentation.Type.PLANTUML,"writing assistant flow");System.out.println("\n=== Writing Assistant UML Flow ===");System.out.println(representation.content());System.out.println("==================================\n");return graph;}// 各个节点实现类...
}

5.4 节点实现类

5.4.1 摘要生成器节点

static class SummarizerNode implements NodeAction {private final ChatClient chatClient;public SummarizerNode(ChatClient chatClient) {this.chatClient = chatClient;}@Overridepublic Map<String, Object> apply(OverAllState state) {String text = (String) state.value("original_text").orElse("");String prompt = "请对以下中文文本进行简洁明了的摘要：\n\n" + text;ChatResponse response = chatClient.prompt(prompt).call().chatResponse();String summary = response.getResult().getOutput().getText();Map<String, Object> result = new HashMap<>();result.put("summary", summary);return result;}
}

5.4.2 反馈分类器节点

static class SummaryFeedbackClassifierNode implements NodeAction {private final ChatClient chatClient;private final String inputKey;public SummaryFeedbackClassifierNode(ChatClient chatClient, String inputKey) {this.chatClient = chatClient;this.inputKey = inputKey;}@Overridepublic Map<String, Object> apply(OverAllState state) {String summary = (String) state.value(inputKey).orElse("");if (!StringUtils.hasText(summary)) {throw new IllegalArgumentException("summary is empty in state");}String prompt = """以下是一个自动生成的中文摘要。请你判断它是否让用户满意。如果满意，请返回 "positive"，否则返回 "negative"：摘要内容：%s""".formatted(summary);ChatResponse response = chatClient.prompt(prompt).call().chatResponse();String output = response.getResult().getOutput().getText();String classification = output.toLowerCase().contains("positive") ? "positive" : "negative";Map<String, Object> updated = new HashMap<>();updated.put("summary_feedback", classification);return updated;}
}

5.4.3 内容改写器节点

static class RewordingNode implements NodeAction {private final ChatClient chatClient;public RewordingNode(ChatClient chatClient) {this.chatClient = chatClient;}@Overridepublic Map<String, Object> apply(OverAllState state) {String summary = (String) state.value("summary").orElse("");String prompt = "请将以下摘要用更优美、生动的语言改写，同时保持信息不变：\n\n" + summary;ChatResponse response = chatClient.prompt(prompt).call().chatResponse();String reworded = response.getResult().getOutput().getText();Map<String, Object> result = new HashMap<>();result.put("reworded", reworded);return result;}
}

5.4.4 标题生成器节点

static class TitleGeneratorNode implements NodeAction {private final ChatClient chatClient;public TitleGeneratorNode(ChatClient chatClient) {this.chatClient = chatClient;}@Overridepublic Map<String, Object> apply(OverAllState state) {String content = (String) state.value("reworded").orElse("");String prompt = "请为以下内容生成一个简洁有吸引力的中文标题：\n\n" + content;ChatResponse response = chatClient.prompt(prompt).call().chatResponse();String title = response.getResult().getOutput().getText();Map<String, Object> result = new HashMap<>();result.put("title", title);return result;}
}

5.4.5 反馈分发器

static class FeedbackDispatcher implements EdgeAction {@Overridepublic String apply(OverAllState state) {String feedback = (String) state.value("summary_feedback").orElse("");if (feedback.contains("positive")) {return "positive";}return "negative";}
}

6. 运行与测试

6.1 启动应用

在项目根目录执行以下命令启动应用：

mvn spring-boot:run

6.2 测试接口

使用浏览器或 API 工具（如 Postman, curl）访问以下接口：

GET http://localhost:8080/write?text=今天我学习了spring-ai-alibaba-graph的相关概念，spring-ai-alibaba-graph做的特别好， 感觉特别开心

6.3 预期响应

接口将返回 JSON 格式的处理结果：

{"summary": "今日学习了spring-ai-alibaba-graph的相关概念，对其优秀表现感到开心。","summary_feedback": "positive","reworded": "今日深入探索了spring-ai-alibaba-graph的相关概念，不禁为其卓越的表现而感到由衷欣喜。这一学习旅程不仅充实了我的知识库，更让我领略到这一技术的独特魅力与无限潜力。","original_text": "今天我学习了spring-ai-alibaba-graph的相关概念，spring-ai-alibaba-graph做的特别好， 感觉特别开心","title": "《深入探索Spring AI Alibaba Graph：一场技术魅力的发现之旅》"
}