本文演示如何通过Microsoft.Extensions.AI和Microsoft.Extensions.VectorData库构建向量搜索系统，主要步骤如下：

1. 为数据集生成嵌入向量
2. 创建并填充向量存储
3. 将用户查询转换为嵌入向量
4. 执行相似度搜索并返回结果

初始化项目

创建控制台应用并安装必要依赖：

dotnet new console -o VectorSearchDemo
cd VectorSearchDemo
dotnet add package Microsoft.Extensions.AI.OpenAI --prerelease
dotnet add package Microsoft.Extensions.VectorData.Abstractions
dotnet add package Microsoft.SemanticKernel.Connectors.InMemory --prerelease
dotnet add package Microsoft.Extensions.Configuration
dotnet add package Microsoft.Extensions.Configuration.UserSecrets
dotnet add package System.Linq.AsyncEnumerable

定义数据模型

创建云服务信息实体类：

using Microsoft.Extensions.VectorData;

namespace VectorSearchDemo;

public class CloudServiceEntry
{
    [VectorStoreKey]
    public int Id { get; set; }

    [VectorStoreData]
    public string ServiceName { get; set; }

    [VectorStoreData]
    public string ServiceDesc { get; set; }

    [VectorStoreVector(
        Dimensions: 384,
        DistanceFunction = DistanceFunction.CosineSimilarity)]
    public ReadOnlyMemory<float> VectorData { get; set; }
}

准备数据集

初始化云服务描述数据：

List<CloudServiceEntry> services = new()
{
    new() {
        Id = 0,
        ServiceName = "Azure App Service",
        ServiceDesc = "托管.NET、Java等应用的完全托管服务，自动处理高可用和扩展"
    },
    new() {
        Id = 1,
        ServiceName = "Azure Service Bus",
        ServiceDesc = "企业级消息代理服务，支持点对点和发布订阅模式"
    },
    new() {
        Id = 2,
        ServiceName = "Azure Blob Storage",
        ServiceDesc = "云文件存储服务，支持海量数据存储和高可用架构"
    }
};

配置嵌入生成器

连接OpenAI服务生成嵌入向量：

var config = new ConfigurationBuilder()
    .AddUserSecrets<Program>()
    .Build();

var embeddingGenerator = new OpenAIClient(
    new ApiKeyCredential(config["OpenAIKey"]))
    .GetEmbeddingClient(model: config["ModelName"])
    .AsIEmbeddingGenerator();

构建向量存储

填充向量数据库：

var vectorDb = new InMemoryVectorStore();
var collection = vectorDb.GetCollection<int, CloudServiceEntry>("cloudServices");
await collection.EnsureCollectionExistsAsync();

foreach (var service in services)
{
    service.VectorData = await embeddingGenerator.GenerateVectorAsync(service.ServiceDesc);
    await collection.UpsertAsync(service);
}

执行向量搜索

转换查询并检索结果：

string query = "存储Word文档的最佳服务是什么？";
var queryVector = await embeddingGenerator.GenerateVectorAsync(query);

var matches = collection.SearchAsync(queryVector, top: 3);
await foreach (var match in matches)
{
    Console.WriteLine($"服务: {match.Record.ServiceName}");
    Console.WriteLine($"匹配度: {match.Score:F4}");
}