ai-search 生成建议问
简介
在现代对话系统中，为了更好地满足用户需求，常常需要根据先前的对话上下文为用户生成一些后续的建议问题（suggestions），帮助用户更深入地与大模型或搜索引擎进行交互。本文将介绍如何基于大模型来生成这些建议问，并通过示例代码演示具体的实现方法和使用流程。
核心思路
预设 Prompt
提前编写一个 Prompt（suggestion_generator_prompt.txt），用以向大模型描述具体的任务目标和输出格式。
该 Prompt 中指定了以下要点：- 需要生成 4-5 条建议问（suggestions）。
- 建议问需与对话上下文紧密相关，且对用户来说具有实用价值。
- 每条建议问需要具备中等长度、信息量适中并能够进一步引导用户与大模型的对话。
- 最终的输出格式需要用特定 XML 标签（
<suggestions>和</suggestions>）将所有建议问包裹起来。
对话上下文输入
在调用大模型进行生成时，会将先前的对话历史（histories）一并传入，让大模型知道目前的对话场景和用户的意图。结果解析
生成完成后，需要对大模型返回的文本结果进行解析，从其中提取<suggestions>标签中的内容（即建议问）。然后，将这些建议问返回或处理后用于后续的对话或界面展示。
代码实现
以下代码示例演示了如何结合大模型接口及预设 Prompt 来生成建议问，并将其应用到 Java 项目中。示例使用了自定义的 OpenAiClient、ChatMessage、ChatResponseVo 等类，分别实现大模型的请求、消息结构封装和响应解析等功能。
1. suggestion_generator_prompt.txt
You are an AI suggestion generator for an AI powered search engine. You will be given a conversation below. You need to generate 4-5 suggestions based on the conversation. The suggestion should be relevant to the conversation that can be used by the user to ask the chat model for more information.
You need to make sure the suggestions are relevant to the conversation and are helpful to the user. Keep a note that the user might use these suggestions to ask a chat model for more information.
Make sure the suggestions are medium in length and are informative and relevant to the conversation.
You must reply using the user's message language.
如果调用大模型进行 json 输出,仅仅需要上半部分提示词即可
Provide these suggestions separated by newlines between the XML tags <suggestions> and </suggestions>. For example:
<suggestions>
Tell me more about SpaceX and their recent projects
What is the latest news on SpaceX?
Who is the CEO of SpaceX?
</suggestions>
Conversation:
该文件中规定了生成建议问所需的上下文，以及输出格式要求。大模型会根据其中的指令来返回相应的建议问。
Google Gemini 模型请求示例
API_KEY="YOUR_API_KEY"
curl \
-X POST https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent?key=${API_KEY} \
-H 'Content-Type: application/json' \
-d @<(echo '{
"contents": [
{
"role": "user",
"parts": [
{
"text": "Hot was the professor Tong Li."
}
]
},
],
"systemInstruction": {
"role": "user",
"parts": [
{
"text": "You are an AI suggestion generator for an AI powered search engine. You will be given a conversation below. You need to generate 4-5 suggestions based on the conversation. The suggestion should be relevant to the conversation that can be used by the user to ask the chat model for more information.\nYou need to make sure the suggestions are relevant to the conversation and are helpful to the user. Keep a note that the user might use these suggestions to ask a chat model for more information.\nMake sure the suggestions are medium in length and are informative and relevant to the conversation."
}
]
},
"generationConfig": {
"temperature": 1,
"topK": 40,
"topP": 0.95,
"maxOutputTokens": 8192,
"responseMimeType": "application/json",
"responseSchema": {
"type": "object",
"properties": {
"suggestions": {
"type": "array",
"items": {
"type": "string"
}
}
}
}
}
}')
3. SearchSuggestionQuesionService 类
该类主要功能是调用大模型接口，并将预设好的 Prompt 与对话上下文拼接后发送给大模型，最终从返回的内容中抽取出建议问。
package com.litongjava.perplexica.services;
import java.util.ArrayList;
import java.util.List;
import com.litongjava.gemini.GeminiChatRequestVo;
import com.litongjava.gemini.GeminiChatResponseVo;
import com.litongjava.gemini.GeminiClient;
import com.litongjava.gemini.GeminiContentVo;
import com.litongjava.gemini.GeminiGenerationConfigVo;
import com.litongjava.gemini.GeminiPartVo;
import com.litongjava.gemini.GeminiResponseSchema;
import com.litongjava.gemini.GeminiSystemInstructionVo;
import com.litongjava.gemini.GoogleGeminiModels;
import com.litongjava.openai.chat.ChatMessage;
import com.litongjava.openai.chat.OpenAiChatMessage;
import com.litongjava.openai.chat.OpenAiChatResponseVo;
import com.litongjava.openai.client.OpenAiClient;
import com.litongjava.openai.consts.OpenAiModels;
import com.litongjava.template.PromptEngine;
import com.litongjava.tio.utils.hutool.StrUtil;
import lombok.extern.slf4j.Slf4j;
@Slf4j
public class SearchSuggestionQuesionService {
private String prompt = PromptEngine.renderToString("suggestion_generator_prompt.txt");
public String generate(List<ChatMessage> histories) {
// 调用大模型，传入模型名称、Prompt 与对话历史
String content = useGemini(prompt, histories);
if(StrUtil.isBlank(content)) {
return null;
}
return content;
}
public String useGemini(String prompt, List<ChatMessage> histories) {
List<GeminiContentVo> contents = new ArrayList<>();
// 1. 如果有对话历史，则构建 role = user / model 的上下文内容
for (int i = 0; i < histories.size(); i++) {
ChatMessage chatMessage = histories.get(i);
String role = chatMessage.getRole();
if ("human".equals(role)) {
role = "user";
} else {
role = "model";
}
contents.add(new GeminiContentVo(role, chatMessage.getContent()));
}
GeminiChatRequestVo reqVo = new GeminiChatRequestVo(contents);
GeminiPartVo geminiPartVo = new GeminiPartVo(prompt);
GeminiSystemInstructionVo geminiSystemInstructionVo = new GeminiSystemInstructionVo();
geminiSystemInstructionVo.setParts(geminiPartVo);
reqVo.setSystem_instruction(geminiSystemInstructionVo);
//
String key = "suggestions";
GeminiResponseSchema schema = GeminiResponseSchema.array(key);
GeminiGenerationConfigVo geminiGenerationConfigVo = new GeminiGenerationConfigVo();
geminiGenerationConfigVo.buildJsonValue().setResponseSchema(schema);
reqVo.setGenerationConfig(geminiGenerationConfigVo);
GeminiChatResponseVo generate = GeminiClient.generate(GoogleGeminiModels.GEMINI_2_0_FLASH_EXP, reqVo);
return generate.getCandidates().get(0).getContent().getParts().get(0).getText();
}
public String useOpenAi(String prompt, List<ChatMessage> histories) {
List<OpenAiChatMessage> messages = new ArrayList<>();
for (ChatMessage item : histories) {
messages.add(new OpenAiChatMessage(item.getRole(), item.getContent()));
}
OpenAiChatResponseVo chatResponseVo = OpenAiClient.chatCompletions(OpenAiModels.GPT_4O_MINI, prompt, messages);
String content = chatResponseVo.getChoices().get(0).getMessage().getContent();
return content;
}
}
- Prompt 读取：使用
Engine读取并渲染suggestion_generator_prompt.txt文件内容。 - 向大模型发送请求：利用
OpenAiClient.chatComplections方法，将模型名称、Prompt 与对话历史一起传递给大模型。 - 解析返回结果：大模型的返回结果会是一个包含若干回复选项的 JSON 对象。在这里通过
getChoices().get(0).getMessage().getContent()获取到第一个回复的正文，然后调用TagUtils.extractSuggestions(content)方法提取<suggestions>标签中的建议问并返回给调用方。
3. SearchSuggestionQuesionServiceTest 测试类
下面演示如何在单元测试中使用 SearchSuggestionQuesionService 来生成建议问。示例通过 histories.add(ChatMessage.buildUser("How was the professor Tong Li")); 的方式模拟用户的一次对话输入，查看大模型给出的建议问输出。
package com.litongjava.perplexica.services;
import java.util.ArrayList;
import java.util.List;
import org.junit.Test;
import com.litongjava.jfinal.aop.Aop;
import com.litongjava.openai.chat.ChatMessage;
import com.litongjava.perplexica.config.EnjoyEngineConfig;
import com.litongjava.tio.utils.environment.EnvUtils;
public class SearchSuggestionQuesionServiceTest {
@Test
public void test() {
// 加载环境变量或配置信息
EnvUtils.load();
// 配置模板引擎
new EnjoyEngineConfig().config();
// 构造一段对话历史
List<ChatMessage> histories = new ArrayList<>();
histories.add(ChatMessage.buildUser("How was the professor Tong Li"));
// 调用 SearchSuggestionQuesionService 进行建议问生成
String generate = Aop.get(SearchSuggestionQuesionService.class).generate(histories);
// 输出结果
System.out.println(generate);
}
}
运行测试后，控制台可能会输出类似以下内容：
What are some notable contributions of Professor Tong Li in their field?
Can you provide a biography or background information on Professor Tong Li?
What are the key research areas that Professor Tong Li specializes in?
What is the impact of Professor Tong Li's work on their academic community?
Are there any famous publications or works by Professor Tong Li that I should know about?
这些建议问有助于用户进一步了解对话主题，或深入挖掘教授的研究方向、成就以及相关文献，从而丰富后续的交互体验。
4. 示例输出解读
通过上述实现，最终我们可以得到一批符合指定格式、与对话内容密切相关的建议问。用户可以基于这些问题向大模型发起新的请求，进行更深入的探讨。此外，开发者或前端也可将这些建议问以合适的方式展示给用户，帮助引导或提示用户进行更有价值的后续对话。
5. Handler
package com.litongjava.perplexica.handler;
import java.util.ArrayList;
import java.util.List;
import com.alibaba.fastjson2.JSONArray;
import com.alibaba.fastjson2.JSONObject;
import com.litongjava.jfinal.aop.Aop;
import com.litongjava.openai.chat.ChatMessage;
import com.litongjava.perplexica.services.SearchSuggestionQuesionService;
import com.litongjava.perplexica.vo.SearchChatMesageVo;
import com.litongjava.perplexica.vo.WebPageSource;
import com.litongjava.tio.boot.http.TioRequestContext;
import com.litongjava.tio.http.common.HttpRequest;
import com.litongjava.tio.http.common.HttpResponse;
import com.litongjava.tio.http.common.MimeType;
import com.litongjava.tio.http.server.util.Resps;
import com.litongjava.tio.utils.hutool.StrUtil;
import com.litongjava.tio.utils.json.FastJson2Utils;
public class SearchSuggestionQuesionHandler {
public HttpResponse index(HttpRequest request) {
String bodyString = request.getBodyString();
HttpResponse response = TioRequestContext.getResponse();
if (StrUtil.isBlank(bodyString)) {
return Resps.fail(response, "request body can not be empty");
}
JSONObject jsonObject = FastJson2Utils.parseObject(bodyString);
JSONArray chatHistory = jsonObject.getJSONArray("chatHistory");
if (chatHistory == null) {
return Resps.fail(response, "chatHistory can not be empty");
}
List<SearchChatMesageVo> searchChatMessages = chatHistory.toJavaList(SearchChatMesageVo.class);
List<ChatMessage> chatMessages = new ArrayList<>();
for (SearchChatMesageVo searchChatMesageVo : searchChatMessages) {
String role = searchChatMesageVo.getRole();
List<WebPageSource> sources = searchChatMesageVo.getSources();
if (sources != null && sources.size() > 0) {
StringBuffer stringBuffer = new StringBuffer();
for (WebPageSource source : sources) {
String pageContent = source.getPageContent();
if (pageContent != null) {
stringBuffer.append("source url:").append(source.getMetadata().getUrl()).append("content:").append(pageContent).append(" \r\n");
}
}
chatMessages.add(new ChatMessage(role, stringBuffer.toString()));
}
String content = searchChatMesageVo.getContent();
chatMessages.add(new ChatMessage(role, content));
}
String generated = Aop.get(SearchSuggestionQuesionService.class).generate(chatMessages);
if (generated != null) {
response.setString(generated, null, MimeType.TEXT_PLAIN_JSON.toString());
}
return response;
}
}
接口
为什么在回顾一下生成建议问的接口
/api/suggestions
payload
{
"chatHistory": [
{
"id": "7138169222103040",
"role": "user",
"content": "When is the spring break in kcpiolani community college?",
"created_at": 1742717158761,
"sources": [
{
"metadata": {
"url": "https://www.kapiolani.hawaii.edu/events/spring-break-2/",
"title": "Spring Break | Kapiolani Community College - kapiolani.hawaii.edu"
},
"pageContent": "18marAll Day 22 Spring Break. Event Details. Spring Break 2024, March 18 - 22. Time. March 18, 2024 - March 22, 2024 (All Day) (GMT-10:00) Calendar GoogleCal. Report a Concern. 4303 Diamond Head Rd Honolulu, HI 96816 (808) 734-9000. An equal opportunity institution accredited by ACCJC. Consumer Info. Follow Us. Follow; Follow ..."
},
{
"metadata": {
"url": "https://www.kapiolani.hawaii.edu/classes/academic-calendar/",
"title": "Academic Calendar | Kapiolani Community College"
},
"pageContent": "Last day to withdraw from full-semester classes with 100% tuition refund* online at www.star.hawaii.edu ... Mar 17-21 - Spring Recess Mar 26 - Prince Kūhiō Day Apr 18 - Good Friday May 26 - Memorial Day Jun 11 - King Kamehameha I Day Jul 04 -Independence Day"
},
{
"metadata": {
"url": "https://www.kapiolani.hawaii.edu/wp-content/uploads/2024-2025_CATALOG_Table_of_Contents.pdf",
"title": "PDF UNIVERSITY OF HAWAII - Kapiʻolani Community College"
},
"pageContent": "www.kapiolani.hawaii.edu . FALL 2024 Academic Calendar (8/21/2024 ... March 17 - 21 Spring Break (Campus closed) March 26 HOLIDAY: Prince Jonah Kūhiō Kalanianaole Day April 18 HOLIDAY: Good Friday May 7 LAST DAY OF INSTRUCTION May 9 - 16 Final Examination Period"
},
{
"metadata": {
"url": "https://www.sis.hawaii.edu/uhdad/avail.classes?i=KAP&t=202530",
"title": "Kapiolani Community College - Spring 2025 - sis.hawaii.edu"
},
"pageContent": "Subjects offered by Kapiolani Community College for Spring 2025: We encourage you to regularly check this schedule of classes and your class schedule in STAR /MyUH, as well as your UH email account and Laulima/Lamakū for additional messages."
},
{
"metadata": {
"url": "https://www.hawaii.edu/academic-calendar/",
"title": "Academic Calendar - University of Hawaii System"
},
"pageContent": "The 10 University of Hawaiʻi campuses begin each semester on the same date and observe the same holidays. See campus academic calendars for dates and deadlines for registration, advising and other academic activities. Summer session dates are those of the UH Mānoa Outreach College."
},
{
"metadata": {
"url": "https://guides.library.kapiolani.hawaii.edu/testing",
"title": "Kapiʻolani CC Library LibGuides: Testing Center: HOME"
},
"pageContent": "The Testing Center will be closed Spring Break, March 17 - 21, 2025. The Testing Center will also be closed Wednesday, March 26, 2025, in observance of Kuhio Day. ... guides.library.kapiolani.hawaii.edu/testing ... 808-734-9453 • kcctest@hawaii.edu."
},
{
"metadata": {
"url": "https://www.sis.hawaii.edu/uhdad/avail.classes?i=KAP&t=202530&s=PHIL",
"title": "Kapiolani Community College - Spring 2025 - PHIL - sis.hawaii.edu"
},
"pageContent": "To take a hybrid class, students must have regular access to a recent model desktop or laptop computer and a reliable Internet connection. Email the instructor (kchen@hawaii.edu) as needed for additional information and check your UH email account regularly for class notifications. This course section will use the Laulima Learning Management ..."
},
{
"metadata": {
"url": "https://guides.library.kapiolani.hawaii.edu/libraryinfo/hours",
"title": "Home - Library Hours - Kapiolani Community College"
},
"pageContent": "Spring Break: CLOSED Online Chat available: 8:30 am - 4:00 pm. March 26, 2025 (Wednesday) Prince Kūhiō Day: CLOSED: April 18, 2025 (Friday) Good Friday: CLOSED: Last Updated: Mar 4, 2025 3:01 PM. Kapiʻolani Community College Library • 4303 Diamond Head Road • Honolulu Hawaiʻi 96816-4496 guides.library.kapiolani.hawaii.edu ..."
},
{
"metadata": {
"url": "https://www.kapiolani.hawaii.edu/wp-content/uploads/Spring-2023-Academic-Calendar.pdf",
"title": "PDF SPRING 2023 ACADEMIC CALENDAR (January 9, 2023 - May 12, 2023)"
},
"pageContent": "SPRING 2023 ACADEMIC CALENDAR (January 9, 2023 - May 12, 2023) Monday, November 7, 2022 Online registration begins for currently enrolled students and returning ... www.star.hawaii.edu Friday, February 3 Last day to apply and register for credit by examination Monday, February 20 HOLIDAY: Presidents Day ..."
},
{
"metadata": {
"url": "https://www.sis.hawaii.edu/uhdad/avail.classes?i=KAP&t=202330",
"title": "Kapiolani Community College - Spring 2023 - University of Hawaiʻi"
},
"pageContent": "Subjects offered by Kapiolani Community College for Spring 2023: Class availability information for this term is not available at this time. ©2025 University of Hawaii. Updated: 03/19/2025 01:46:35 AM HST ..."
}
]
},
{
"id": "493422523526352896",
"role": "assistant",
"content": "Spring break at Kapiolani Community College (KCC) occurs in mid-March [1]. In 2024, it was from March 18-22 [1]. For 2025, spring break is scheduled for March 17-21, during which the campus will be closed [3]. Additionally, the Testing Center will be closed during this time [6]. The KCC Library will also be closed for spring break, but online chat support will be available from 8:30 am to 4:00 pm [8]. KCC observes the same holidays as the other University of Hawaii campuses [5].\\n",
"created_at": 1742717161239,
"sources": null
}
],
"chatModel": {
"provider": "openai",
"model": "gpt-3.5-turbo"
}
}
resposne
{
"suggestions": [
"What are the registration deadlines for Kapiʻolani Community College?",
"Can you provide more information about the tuition fees at Kapiʻolani Community College?",
"Are there any important dates or events to be aware of at Kapiʻolani Community College?",
"What are the academic calendar dates for Kapiʻolani Community College?"
]
}
总结
在对话系统或搜索系统中生成建议问的关键要点是：上下文理解 + 明确的任务指令 + 预期格式化输出。
通过预先编写并封装一个合适的 Prompt，将对话历史和格式化要求一起传递给大模型，再配合解析流程从文本中提取最终结果，就能轻松实现“自动生成建议问”这一功能。此做法既能提升用户体验，也为进一步挖掘用户需求与提供个性化服务奠定了基础。