{"componentChunkName":"component---src-templates-acg-portal-new-template-tsx","path":"/pmtwm0iu0","result":{"data":{"markdownRemark":{"html":"

本文介绍如何在百度胜算中使用图文搜索能力，包括创建分析与 AI 搜索实例、连接 Notebook、创建向量表、生成多模态 embedding，以及执行图搜图和文搜图。

\n

前提条件

\n\n

操作步骤

\n

步骤一：创建 Notebook

\n

前往工作台，选择目标项目，依次单击右上角创建>数据处理>Notebook。

\n

\"image

\n

步骤二：连接分析与 AI 搜索实例

\n

进入创建的 Notebook，在右上角下拉框中选择处于运行中的分析与 AI 搜索实例。此时可以在 Notebook 中按本文顺序执行 SQL，体验图文搜索。

\n

\"image

\n

步骤三：创建测试数据

\n

以下示例使用 rdtest 数据库和 test_table 数据表。

\n

创建数据库

\n

执行以下 SQL 创建数据库：

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1CREATE DATABASE IF NOT EXISTS rdtest;
\n
\n
\n
\n \n

\"image

\n

选择数据库

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1USE rdtest;
\n
\n
\n
\n \n

\"image

\n

创建数据表和 ANN 索引

\n

执行以下 SQL 创建数据表。embedding 列用于保存图片或文本的向量，ANN 索引使用 HNSW 算法、L2 距离，向量维度为 1536。

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1CREATE TABLE IF NOT EXISTS `test_table` (\n2  `id` INT NOT NULL COMMENT "数据唯一编号",\n3  `content` varchar(1000) NULL COMMENT "内容说明",\n4  `path` varchar(1000) NULL COMMENT "图片路径",\n5  `embedding` array<float> COMMENT "图片向量数据",\n6  INDEX index_ann (`embedding`) USING ANN PROPERTIES(\n7    "distance" = "l2",\n8    "dimension" = "1536",\n9    "algorithm" = "hnsw"\n10  )\n11) ENGINE=OLAP\n12DUPLICATE KEY(`id`)\n13DISTRIBUTED BY HASH(`id`) BUCKETS 2;
\n
\n
\n
\n \n

\"image

\n

导入图片数据并生成 embedding

\n
\n

如果测试环境暂时无法搭建图片服务，可以跳过图片数据导入，改用使用文本embedding进行功能验证。

\n
\n

执行以下 SQL 导入测试数据，并通过 embedding() 函数生成图片向量：

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1INSERT INTO test_table (id, content, path, embedding) VALUES\n2    (1, 'xiezi1', 'http://xxx.bj.bcebos.com/dev/wgf/images/xiezi1.jpg',\n3        embedding('http://xxx.bj.bcebos.com/dev/wgf/images/xiezi1.jpg', 'image', 'multimodal_embedding_server')),\n4    (2, 'xiezi2', 'http://xxx.bj.bcebos.com/dev/wgf/images/xiezi2.png',\n5        embedding('http://xxx.bj.bcebos.com/dev/wgf/images/xiezi2.png', 'image', 'multimodal_embedding_server')),\n6    (3, 'xiezi3', 'http://xxx.bj.bcebos.com/dev/wgf/images/xiezi3.jpg',\n7        embedding('http://xxx.bj.bcebos.com/dev/wgf/images/xiezi3.jpg', 'image', 'multimodal_embedding_server')),\n8    (4, 'xiezi_test1', 'http://xxx.bj.bcebos.com/dev/wgf/images/xiezi_test1.jpg',\n9        embedding('http://xxx.bj.bcebos.com/dev/wgf/images/xiezi_test1.jpg', 'image', 'multimodal_embedding_server')),\n10    (5, 'mao1', 'http://xxx.bj.bcebos.com/dev/wgf/images/mao1.jpg',\n11        embedding('http://xxx.bj.bcebos.com/dev/wgf/images/mao1.jpg', 'image', 'multimodal_embedding_server')),\n12    (6, 'mao2', 'http://xxx.bj.bcebos.com/dev/wgf/images/mao2.jpg',\n13        embedding('http://xxx.bj.bcebos.com/dev/wgf/images/mao2.jpg', 'image', 'multimodal_embedding_server')),\n14    (7, 'mao_test1', 'http://xxx.bj.bcebos.com/dev/wgf/images/mao_test1.jpg',\n15        embedding('http://xxx.bj.bcebos.com/dev/wgf/images/mao_test1.jpg', 'image', 'multimodal_embedding_server'));
\n
\n
\n
\n \n\n

查看导入的数据

\n

执行以下 SQL 查看已导入的数据：

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, path\n2FROM test_table\n3LIMIT 10;
\n
\n
\n
\n \n

\"image

\n

执行图文搜索

\n

以下示例均使用 test_table 中的 embedding 列进行检索，并返回 __DISTANCE 距离列。通常情况下，距离越小表示结果越相似。

\n

TopK检索（图搜图：搜索相似鞋子）

\n

将鞋子图片 URL 生成查询向量，返回与该图片最相似的图片。

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, path, __DISTANCE\n2FROM test_table\n3ORDER BY ann_distance(\n4    embedding,\n5    embedding('http://xxx.bj.bcebos.com/dev/wgf/images/xiezi_test1.jpg', 'image', 'multimodal_embedding_server')\n6)\n7LIMIT 10;
\n
\n
\n
\n \n

\"image

\n

TopK检索（图搜图：搜索相似猫）

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, path, __DISTANCE\n2FROM test_table\n3ORDER BY ann_distance(\n4    embedding,\n5    embedding('http://xxx.bj.bcebos.com/dev/wgf/images/mao_test1.jpg', 'image', 'multimodal_embedding_server')\n6)\n7LIMIT 10;
\n
\n
\n
\n \n

\"image

\n

范围检索（图搜图：按距离范围搜索鞋子）

\n

只返回距离在 0 到 1.5 之间的图片，并按距离升序排列。

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, path, __DISTANCE\n2FROM test_table\n3WHERE ann_distance(\n4    embedding,\n5    embedding('http://xxx.bj.bcebos.com/dev/wgf/images/xiezi_test1.jpg', 'image', 'multimodal_embedding_server')\n6) BETWEEN 0 AND 1.5\n7ORDER BY __DISTANCE;
\n
\n
\n
\n \n

\"image

\n

范围检索：（图搜图：按距离范围搜索猫）

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, path, __DISTANCE\n2FROM test_table\n3WHERE ann_distance(\n4    embedding,\n5    embedding('http://xxx.bj.bcebos.com/dev/wgf/images/mao_test1.jpg', 'image', 'multimodal_embedding_server')\n6) BETWEEN 0 AND 1\n7ORDER BY __DISTANCE;
\n
\n
\n
\n \n

\"image

\n

TopK检索（文搜图：搜索猫）

\n

将文本“猫咪相关的图片”生成查询向量，搜索相似图片。

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, path, __DISTANCE\n2FROM test_table\n3ORDER BY ann_distance(\n4    embedding,\n5    embedding('猫咪相关的图片', 'text', 'multimodal_embedding_server')\n6)\n7LIMIT 10;
\n
\n
\n
\n \n

\"image

\n

TopK检索（文搜图：使用更具体的描述搜索猫）

\n

通过增加“站着的”等修饰词，使查询描述更具体。

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, path, __DISTANCE\n2FROM test_table\n3ORDER BY ann_distance(\n4    embedding,\n5    embedding('站着的猫咪图片', 'text', 'multimodal_embedding_server')\n6)\n7LIMIT 10;
\n
\n
\n
\n \n

\"image

\n

TopK检索（文搜图：搜索鞋子）

\n

使用文本查询“皮鞋相关的图片”，并限制距离范围。

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, __DISTANCE\n2FROM test_table\n3WHERE ann_distance(\n4    embedding,\n5    embedding('皮鞋相关的图片', 'text', 'multimodal_embedding_server')\n6) BETWEEN 0 AND 1.5\n7ORDER BY __DISTANCE\n8LIMIT 10;
\n
\n
\n
\n \n

\"image

\n

范围检索（文搜图：按距离范围搜索鞋子）

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, __DISTANCE\n2FROM test_table\n3WHERE ann_distance(\n4    embedding,\n5    embedding('鞋子相关的图片', 'text', 'multimodal_embedding_server')\n6) BETWEEN 0 AND 1.5\n7ORDER BY __DISTANCE\n8LIMIT 10;
\n
\n
\n
\n \n

\"image

\n

单独调用 embedding() 函数

\n

可以直接调用 embedding() 函数生成向量：

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT embedding(\n2    'http://xxx.bj.bcebos.com/dev/wgf/images/xiezi_test1.jpg',\n3    'image',\n4    'multimodal_embedding_server'\n5) AS embedding;
\n
\n
\n
\n \n

\"image

\n

使用文本生成 embedding

\n

如果测试环境无法提供图片 URL，可以将文本转换为 embedding，使用同一张表验证文本检索图片的流程。

\n

导入文本数据

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1INSERT INTO test_table (id, content, path, embedding) VALUES\n2    (1, 'xiezi1', 'xiezi1',\n3        embedding('一双棕色的鞋子', 'text', 'multimodal_embedding_server')),\n4    (2, 'xiezi2', 'xiezi2',\n5        embedding('一双红色的鞋子', 'text', 'multimodal_embedding_server')),\n6    (6, 'mao2', 'mao2',\n7        embedding('一只黄色的猫', 'text', 'multimodal_embedding_server'));
\n
\n
\n
\n \n

查看文本数据

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, path\n2FROM test_table\n3LIMIT 10;
\n
\n
\n
\n \n

执行文本 Top-K 查询

\n

使用文本查询生成 embedding，并返回最相似的图片记录：

\n\n
\n
\n
\n SQL\n \n
\n
\n 
1SELECT id, content, path, __DISTANCE\n2FROM test_table\n3ORDER BY ann_distance(\n4    embedding,\n5    embedding('shoes', 'text', 'multimodal_embedding_server')\n6)\n7LIMIT 10;\n8\n9SELECT id, content, path, __DISTANCE\n10FROM test_table\n11ORDER BY ann_distance(\n12    embedding,\n13    embedding('cat', 'text', 'multimodal_embedding_server')\n14)\n15LIMIT 10;
\n
\n
\n
\n \n

执行结果参考下图：

\n

\"1dfee54f58f86e82dc91120a1c44687f.png\"

","fields":{"slug":"pmtwm0iu0","title":"图文搜索最佳实践","date":"2026-09-11","extractedHeadings":[]},"headings":[{"value":"前提条件","depth":2},{"value":"操作步骤","depth":2},{"value":"步骤一：创建 Notebook","depth":3},{"value":"步骤二：连接分析与 AI 搜索实例","depth":3},{"value":"步骤三：创建测试数据","depth":2},{"value":"创建数据库","depth":3},{"value":"选择数据库","depth":3},{"value":"创建数据表和 ANN 索引","depth":3},{"value":"导入图片数据并生成 embedding","depth":3},{"value":"查看导入的数据","depth":2},{"value":"执行图文搜索","depth":2},{"value":"TopK检索（图搜图：搜索相似鞋子）","depth":3},{"value":"TopK检索（图搜图：搜索相似猫）","depth":3},{"value":"范围检索（图搜图：按距离范围搜索鞋子）","depth":3},{"value":"范围检索：（图搜图：按距离范围搜索猫）","depth":3},{"value":"TopK检索（文搜图：搜索猫）","depth":3},{"value":"TopK检索（文搜图：使用更具体的描述搜索猫）","depth":3},{"value":"TopK检索（文搜图：搜索鞋子）","depth":3},{"value":"范围检索（文搜图：按距离范围搜索鞋子）","depth":3},{"value":"单独调用 embedding() 函数","depth":3},{"value":"使用文本生成 embedding","depth":2},{"value":"导入文本数据","depth":3},{"value":"查看文本数据","depth":3},{"value":"执行文本 Top-K 查询","depth":3}]}},"pageContext":{"isCreatedByStatefulCreatePages":false,"slug":"pmtwm0iu0","prev":{"id":"jmtwlxocl","name":"向量检索性能测试报告","path":"jmtwlxocl","filePath":"百度胜算Doris增强介绍/向量检索/向量检索性能测试报告.md","seo":null,"parentIds":["4mtwk96ql","mmtwk9l0q"],"parents":[{"id":"4mtwk96ql","documentId":"a40119a5-3e8a-43bb-bee7-52e27df99e0e","name":"百度胜算Doris增强介绍","repoName":"DataBuilder","filePath":"百度胜算Doris增强介绍","disabled":false,"path":"4mtwk96ql","lastMergeTime":null,"isApiDoc":null,"httpMethod":null,"seo":null,"sourceOrgName":null,"sourceRepoName":null,"sourceDocumentId":"","solutionTag":null,"solutionColor":null},{"id":"mmtwk9l0q","documentId":"d6e96a6f-6485-48ba-a8cb-62b53b8c4583","name":"向量检索","repoName":"DataBuilder","filePath":"百度胜算Doris增强介绍/向量检索","disabled":false,"path":"mmtwk9l0q","lastMergeTime":null,"isApiDoc":null,"httpMethod":null,"seo":null,"sourceOrgName":null,"sourceRepoName":null,"sourceDocumentId":"","solutionTag":null,"solutionColor":null}]},"next":{"id":"kmtwkpnmh","name":"全文检索","path":"kmtwkpnmh","filePath":"百度胜算Doris增强介绍/全文检索/全文检索相关性使用手册.md","seo":null,"parentIds":["4mtwk96ql","Ymtwk9t9p"],"parents":[{"id":"4mtwk96ql","documentId":"a40119a5-3e8a-43bb-bee7-52e27df99e0e","name":"百度胜算Doris增强介绍","repoName":"DataBuilder","filePath":"百度胜算Doris增强介绍","disabled":false,"path":"4mtwk96ql","lastMergeTime":null,"isApiDoc":null,"httpMethod":null,"seo":null,"sourceOrgName":null,"sourceRepoName":null,"sourceDocumentId":"","solutionTag":null,"solutionColor":null},{"id":"Ymtwk9t9p","documentId":"f4f6d5e1-fca7-469c-89c9-1a3535cc9e5c","name":"全文检索","repoName":"DataBuilder","filePath":"百度胜算Doris增强介绍/全文检索","disabled":false,"path":"Ymtwk9t9p","lastMergeTime":null,"isApiDoc":null,"httpMethod":null,"seo":null,"sourceOrgName":null,"sourceRepoName":null,"sourceDocumentId":"","solutionTag":null,"solutionColor":null}]},"parents":[{"id":"4mtwk96ql","documentId":"a40119a5-3e8a-43bb-bee7-52e27df99e0e","name":"百度胜算Doris增强介绍","repoName":"DataBuilder","filePath":"百度胜算Doris增强介绍","disabled":false,"path":"4mtwk96ql","lastMergeTime":null,"isApiDoc":null,"httpMethod":null,"seo":null,"sourceOrgName":null,"sourceRepoName":null,"sourceDocumentId":"","solutionTag":null,"solutionColor":null},{"id":"mmtwk9l0q","documentId":"d6e96a6f-6485-48ba-a8cb-62b53b8c4583","name":"向量检索","repoName":"DataBuilder","filePath":"百度胜算Doris增强介绍/向量检索","disabled":false,"path":"mmtwk9l0q","lastMergeTime":null,"isApiDoc":null,"httpMethod":null,"seo":null,"sourceOrgName":null,"sourceRepoName":null,"sourceDocumentId":"","solutionTag":null,"solutionColor":null}],"specificSeo":null}}}