![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802010102-6a6e96ce2e149.webp\")
<\/p>\n\u8fd9\u7bc7\u6587\u7ae0\u662f\u7ee7\u4e4b\u524d\u6587\u7ae0 \u201cElasticsearch：\u641c\u7d22\u6559\u7a0b – \u5168\u6587\u641c\u7d22（\u4e00）\u201d\u3002\u5728\u7ee7\u7eed\u8fd9\u7bc7\u6587\u7ae0\u4e4b\u524d，\u4f60\u5fc5\u987b\u4ed4\u7ec6\u9605\u8bfb\u4e4b\u524d\u7684\u90a3\u7bc7\u6587\u7ae0\u3002<\/p>\n<\/p>\n
\u5728 Machine Learning \u4e2d，embedding \u662f\u4e00\u4e2a vector（\u6570\u5b57\u6570\u7ec4），\u7528\u4e8e\u8868\u793a\u73b0\u5b9e\u4e16\u754c\u4e2d\u7684\u5bf9\u8c61，\u4f8b\u5982 words\u3001sentences\u3001images \u6216 videos\u3002<\/p>\n
embedding \u7684\u4e00\u4e2a\u91cd\u8981\u7279\u6027\u662f：\u5982\u679c\u4e24\u4e2a embeddings \u8868\u793a\u7684\u662f\u76f8\u4f3c\u6216\u76f8\u5173\u7684\u73b0\u5b9e\u4e16\u754c\u5b9e\u4f53，\u90a3\u4e48\u5b83\u4eec\u672c\u8eab\u4e5f\u4f1a\u5177\u6709\u4e00\u5b9a\u7684\u76f8\u4f3c\u6027\u3002\u56e0\u6b64，\u53ef\u4ee5\u5bf9 embeddings \u8fdb\u884c\u6bd4\u8f83，\u5e76\u8ba1\u7b97\u5b83\u4eec\u4e4b\u95f4\u7684\u8ddd\u79bb\u3002<\/p>\n
\u4ece search application \u7684\u89d2\u5ea6\u6765\u770b，\u5728 vector space \u4e2d\u641c\u7d22 embeddings，\u901a\u5e38\u80fd\u591f\u627e\u5230\u4e0e\u6982\u5ff5\u76f8\u5173\u7684\u7ed3\u679c，\u800c\u4e0d\u4ec5\u4ec5\u662f\u4e0e\u641c\u7d22\u63d0\u793a\u4e2d\u8f93\u5165\u7684\u7cbe\u786e keywords \u5339\u914d\u7684\u7ed3\u679c\u3002<\/p>\n
\u5728\u672c\u6559\u7a0b\u7684\u8fd9\u4e00\u90e8\u5206\u4e2d，\u4f60\u5c06\u5b66\u4e60\u5982\u4f55\u4f7f\u7528\u514d\u8d39\u63d0\u4f9b\u7684 machine learning models \u751f\u6210 embeddings，\u7136\u540e\u4f7f\u7528 Elasticsearch \u7684 vector database \u529f\u80fd\u6765\u5b58\u50a8\u548c\u641c\u7d22\u8fd9\u4e9b embeddings\u3002<\/p>\n
\u5728\u672c\u90e8\u5206\u7684\u6700\u540e，\u4f60\u8fd8\u5c06\u5b66\u4e60\u5982\u4f55\u5c06 vector search \u4e0e full-text search \u7684\u7ed3\u679c\u7ed3\u5408\u8d77\u6765，\u6784\u5efa\u4e00\u4e2a\u529f\u80fd\u5f3a\u5927\u7684 hybrid search \u89e3\u51b3\u65b9\u6848，\u4ece\u800c\u517c\u987e\u8fd9\u4e24\u79cd\u65b9\u6cd5\u7684\u4f18\u52bf\u3002<\/p>\n<\/p>\n
\u5728\u672c\u8282\u4e2d，\u4f60\u5c06\u5b66\u4e60\u76ee\u524d\u7528\u4e8e\u751f\u6210 text embeddings \u6700\u65b9\u4fbf\u7684\u65b9\u6cd5\u4e4b\u4e00，\u5b83\u57fa\u4e8e SentenceTransformers framework\u3002<\/p>\n
\u5728\u4f60\u63a2\u7d22\u5e76\u719f\u6089 embeddings \u7684\u4f7f\u7528\u8fc7\u7a0b\u4e2d，\u63a8\u8350\u4f7f\u7528 SentenceTransformers\u3002\u8be5 framework \u63d0\u4f9b\u7684 models \u53ef\u4ee5\u5b89\u88c5\u5230\u4f60\u7684\u8ba1\u7b97\u673a\u4e0a，\u65e0\u9700 GPU \u4e5f\u80fd\u83b7\u5f97\u4e0d\u9519\u7684\u6027\u80fd，\u5e76\u4e14\u53ef\u4ee5\u514d\u8d39\u4f7f\u7528\u3002<\/p>\n<\/p>\n
SentenceTransformers framework \u4ee5 Python package \u7684\u5f62\u5f0f\u63d0\u4f9b\u3002<\/p>\n
\u8bf7\u786e\u4fdd\u4f60\u7684 Python virtual environment \u5df2\u6fc0\u6d3b，\u7136\u540e\u5728 terminal \u4e2d\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u5b89\u88c5\u8be5 framework：<\/p>\n
pip install sentence-transformers <\/p>\n
\u4e0e\u5f80\u5e38\u4e00\u6837，\u6bcf\u5f53\u4f60\u5411\u9879\u76ee\u6dfb\u52a0\u65b0\u7684 dependencies \u65f6，\u6700\u597d\u66f4\u65b0\u4f60\u7684 requirements file：<\/p>\n
pip freeze > requirements.txt <\/p>\n
\u4e0b\u4e00\u9879\u4efb\u52a1\u662f\u51b3\u5b9a\u4f7f\u7528\u54ea\u4e2a machine learning model \u6765\u751f\u6210 embeddings\u3002<\/p>\n
documentation \u4e2d\u63d0\u4f9b\u4e86\u4e00\u4efd pretrained models \u5217\u8868\u3002\u7531\u4e8e SentenceTransformers \u662f\u4e00\u4e2a\u975e\u5e38\u6d41\u884c\u7684 framework，\u56e0\u6b64\u9664\u4e86\u5b98\u65b9\u63d0\u4f9b\u7684 models \u5916，\u8fd8\u6709\u8bb8\u591a\u7531\u5176\u4ed6\u7814\u7a76\u4eba\u5458\u521b\u5efa\u7684\u517c\u5bb9 models\u3002\u8981\u67e5\u770b\u6240\u6709\u53ef\u7528\u7684 models，\u53ef\u4ee5\u8bbf\u95ee HuggingFace \u4e0a\u5e26\u6709 SentenceTransformers \u6807\u7b7e\u7684 models\u3002<\/p>\n
\u5bf9\u4e8e\u672c\u6559\u7a0b\u6765\u8bf4，\u4e0d\u9700\u8981\u8fc7\u5ea6\u7ea0\u7ed3 model \u7684\u9009\u62e9，\u56e0\u4e3a\u4efb\u4f55 model \u90fd\u80fd\u591f\u6ee1\u8db3\u9700\u6c42\u3002<\/p>\n
SentenceTransformers documentation \u5bf9\u5176 pretrained models \u7ed9\u51fa\u4e86\u5982\u4e0b\u8bf4\u660e：<\/p>\n
\u201call-* models \u4f7f\u7528\u6240\u6709\u53ef\u7528\u7684\u8bad\u7ec3\u6570\u636e（\u8d85\u8fc7 10 \u4ebf\u4e2a training pairs）\u8fdb\u884c\u8bad\u7ec3，\u65e8\u5728\u4f5c\u4e3a\u901a\u7528\u6a21\u578b\u3002all-mpnet-base-v2 model \u63d0\u4f9b\u6700\u4f73\u8d28\u91cf，\u800c all-MiniLM-L6-v2 \u7684\u901f\u5ea6\u5feb 5 \u500d，\u540c\u65f6\u4ecd\u80fd\u63d0\u4f9b\u826f\u597d\u7684\u8d28\u91cf\u3002\u201d<\/p>\n
\u8fd9\u6bb5\u8bf4\u660e\u8868\u660e，all-MiniLM-L6-v2 model \u5728\u901f\u5ea6\u548c\u8d28\u91cf\u4e4b\u95f4\u53d6\u5f97\u4e86\u4e0d\u9519\u7684\u5e73\u8861，\u56e0\u6b64\u6211\u4eec\u5c06\u4f7f\u7528\u8fd9\u4e2a model\u3002<\/p>\n
\u8bf7\u5728\u8868\u683c\u4e2d\u627e\u5230\u8fd9\u4e2a model，\u7136\u540e\u70b9\u51fb \u201cinfo\u201d \u56fe\u6807\u67e5\u770b\u5b83\u7684\u76f8\u5173\u4fe1\u606f\u3002<\/p>\n
<\/p>\n
\u5173\u4e8e\u6240\u9009 model，\u8fd8\u6709\u4e00\u4e2a\u503c\u5f97\u4e86\u89e3\u7684\u91cd\u8981\u7ec6\u8282，\u90a3\u5c31\u662f\u5b83\u751f\u6210\u7684 embeddings \u7684\u957f\u5ea6，\u6362\u53e5\u8bdd\u8bf4，\u5c31\u662f\u751f\u6210\u7684 vectors \u5305\u542b\u591a\u5c11\u4e2a\u6570\u5b57（dimensions）\u3002<\/p>\n
\u8fd9\u4e00\u70b9\u5f88\u91cd\u8981，\u56e0\u4e3a\u5b83\u4f1a\u76f4\u63a5\u5f71\u54cd\u6240\u9700\u7684\u5b58\u50a8\u7a7a\u95f4\u3002<\/p>\n
\u5bf9\u4e8e all-MiniLM-L6-v2，\u751f\u6210\u7684 vectors \u5177\u6709 384 \u4e2a dimensions\u3002<\/p>\n<\/p>\n
\u4e0b\u9762\u7684 Python \u4ee3\u7801\u6f14\u793a\u4e86\u5982\u4f55\u52a0\u8f7d\u8be5 model\u3002\u4f60\u53ef\u4ee5\u5728 Python shell \u4e2d\u5c1d\u8bd5\u8fd0\u884c\u8fd9\u6bb5\u4ee3\u7801\u3002<\/p>\n
from sentence_transformers import SentenceTransformer
\nmodel = SentenceTransformer('all-MiniLM-L6-v2') <\/p>\n
\u7b2c\u4e00\u6b21\u6267\u884c\u6b64\u64cd\u4f5c\u65f6，\u6a21\u578b\u4f1a\u88ab\u4e0b\u8f7d\u5e76\u5b89\u88c5\u5230\u4f60\u7684\u865a\u62df\u73af\u5883\u4e2d，\u56e0\u6b64\u8c03\u7528\u53ef\u80fd\u9700\u8981\u4e00\u4e9b\u65f6\u95f4\u624d\u80fd\u8fd4\u56de\u3002\u4e00\u65e6\u6a21\u578b\u5b89\u88c5\u5b8c\u6210，\u5b9e\u4f8b\u5316\u5b83\u5e94\u8be5\u4e0d\u4f1a\u82b1\u8d39\u592a\u957f\u65f6\u95f4\u3002<\/p>\n<\/p>\n
\u6a21\u578b\u5b9e\u4f8b\u5316\u540e，\u4f60\u73b0\u5728\u5df2\u7ecf\u53ef\u4ee5\u751f\u6210\u4e00\u4e2a embedding\u3002\u8981\u6267\u884c\u6b64\u64cd\u4f5c，\u5c06\u6e90\u6587\u672c\u4f20\u9012\u7ed9\u6a21\u578b.encode() \u65b9\u6cd5：<\/p>\n
embedding = model.encode('The quick brown fox jumps over the lazy dog') <\/p>\n
\u7ed3\u679c\u662f\u4e00\u4e2a\u5305\u542b\u7ec4\u6210\u8be5 embedding \u7684\u6240\u6709\u6570\u5b57\u7684\u6570\u7ec4\u3002\u6b63\u5982\u4f60\u6240\u8bb0\u5f97\u7684，\u6240\u9009\u6a21\u578b\u751f\u6210\u7684 embeddings \u5177\u6709 384 \u4e2a\u7ef4\u5ea6，\u56e0\u6b64\u8fd9\u5c31\u662f embedding \u6570\u7ec4\u7684\u957f\u5ea6\u3002<\/p>\n<\/p>\n
Elasticsearch \u5b8c\u5168\u652f\u6301\u5b58\u50a8\u548c\u68c0\u7d22\u5411\u91cf，\u8fd9\u4f7f\u5176\u6210\u4e3a\u5904\u7406 embeddings \u7684\u7406\u60f3\u6570\u636e\u5e93\u3002<\/p>\n<\/p>\n
\u5728\u672c\u6559\u7a0b\u7684\u5168\u6587\u641c\u7d22\u7ae0\u8282\u4e2d，\u4f60\u5df2\u7ecf\u5b66\u4e60\u4e86\u5982\u4f55\u521b\u5efa\u4e00\u4e2a\u5305\u542b\u591a\u4e2a\u5b57\u6bb5\u7684\u7d22\u5f15\u3002\u5f53\u65f6\u63d0\u5230，Elasticsearch \u5927\u591a\u6570\u60c5\u51b5\u4e0b\u53ef\u4ee5\u6839\u636e\u6570\u636e\u672c\u8eab\u81ea\u52a8\u786e\u5b9a\u6bcf\u4e2a\u5b57\u6bb5\u6700\u9002\u5408\u4f7f\u7528\u7684\u7c7b\u578b\u3002\u5c3d\u7ba1 Elasticsearch 8.11 \u80fd\u591f\u81ea\u52a8\u6620\u5c04\u67d0\u4e9b\u5411\u91cf\u7c7b\u578b，\u4f46\u5728\u672c\u7ae0\u8282\u4e2d，\u4f60\u5c06\u663e\u5f0f\u5b9a\u4e49\u6b64\u7c7b\u578b，\u4ee5\u6b64\u4f5c\u4e3a\u5b66\u4e60 Elasticsearch \u4e2d\u7c7b\u578b\u6620\u5c04\u66f4\u591a\u77e5\u8bc6\u7684\u673a\u4f1a\u3002<\/p>\n<\/p>\n
\u7d22\u5f15\u4e2d\u6bcf\u4e2a\u5b57\u6bb5\u5173\u8054\u7684\u7c7b\u578b\u662f\u901a\u8fc7\u4e00\u4e2a\u79f0\u4e3a\u6620\u5c04（mapping）\u7684\u8fc7\u7a0b\u786e\u5b9a\u7684，\u8be5\u8fc7\u7a0b\u53ef\u4ee5\u662f\u52a8\u6001\u7684，\u4e5f\u53ef\u4ee5\u662f\u663e\u5f0f\u7684\u3002\u672c\u6559\u7a0b\u5168\u6587\u641c\u7d22\u90e8\u5206\u521b\u5efa\u7684\u6620\u5c04\u90fd\u662f\u7531 Elasticsearch \u52a8\u6001\u751f\u6210\u7684\u3002<\/p>\n
Elasticsearch \u5ba2\u6237\u7aef\u63d0\u4f9b\u4e86\u4e00\u4e2a get_mapping \u65b9\u6cd5，\u5b83\u4f1a\u8fd4\u56de\u67d0\u4e2a\u7ed9\u5b9a\u7d22\u5f15\u5f53\u524d\u751f\u6548\u7684\u7c7b\u578b\u6620\u5c04\u3002\u5982\u679c\u4f60\u60f3\u81ea\u884c\u63a2\u7d22\u8fd9\u4e9b\u6620\u5c04，\u8bf7\u542f\u52a8\u4e00\u4e2a Python shell，\u5e76\u8f93\u5165\u4ee5\u4e0b\u4ee3\u7801：<\/p>\n
from app import es
\nes.es.indices.get_mapping(index='my_documents') <\/p>\n
get_mapping() \u65b9\u6cd5\u8fd4\u56de\u7684\u54cd\u5e94\u662f\u4e00\u4e2a\u5b57\u5178，\u5176\u4e2d\u5305\u542b\u7d22\u5f15\u4e2d\u6bcf\u4e2a\u5b57\u6bb5\u7684\u4fe1\u606f\u3002\u4e3a\u4e86\u65b9\u4fbf\u8d77\u89c1，\u4e0b\u9762\u662f\u672c\u6559\u7a0b\u5168\u6587\u641c\u7d22\u90e8\u5206\u521b\u5efa\u7684 my_documents \u7d22\u5f15\u4e2d\u8fd9\u4e9b\u4fe1\u606f\u7684\u683c\u5f0f\u5316\u7ed3\u6784：<\/p>\n
{
\n "my_documents": {
\n "mappings": {
\n "properties": {
\n "category": {
\n "type": "text",
\n "fields": {
\n "keyword": {
\n "type": "keyword",
\n "ignore_above": 256
\n }
\n }
\n },
\n "content": {
\n "type": "text",
\n "fields": {
\n "keyword": {
\n "type": "keyword",
\n "ignore_above": 256
\n }
\n }
\n },
\n "created_on": {
\n "type": "date"
\n },
\n "name": {
\n "type": "text",
\n "fields": {
\n "keyword": {
\n "type": "keyword",
\n "ignore_above": 256
\n }
\n }
\n },
\n "rolePermissions": {
\n "type": "text",
\n "fields": {
\n "keyword": {
\n "type": "keyword",
\n "ignore_above": 256
\n }
\n }
\n },
\n "summary": {
\n "type": "text",
\n "fields": {
\n "keyword": {
\n "type": "keyword",
\n "ignore_above": 256
\n }
\n }
\n },
\n "updated_at": {
\n "type": "date"
\n },
\n "url": {
\n "type": "text",
\n "fields": {
\n "keyword": {
\n "type": "keyword",
\n "ignore_above": 256
\n }
\n }
\n }
\n }
\n }
\n }
\n}<\/p>\n
\u4ece\u8fd9\u91cc\u4f60\u53ef\u4ee5\u770b\u5230，created_on \u548c updated_at \u5b57\u6bb5\u88ab\u81ea\u52a8\u8bbe\u7f6e\u4e3a date \u7c7b\u578b，\u800c\u5176\u4ed6\u6240\u6709\u5b57\u6bb5\u90fd\u88ab\u8bbe\u7f6e\u4e3a text \u7c7b\u578b\u3002\u5728\u5c1d\u8bd5\u51b3\u5b9a\u7c7b\u578b\u65f6，Elasticsearch \u9996\u5148\u68c0\u67e5\u6570\u636e\u7684\u7c7b\u578b，\u8fd9\u6709\u52a9\u4e8e\u5b83\u4e3a\u5b57\u6bb5\u5206\u914d numeric\u3001boolean \u548c object \u7c7b\u578b\u3002\u5f53\u5b57\u6bb5\u6570\u636e\u662f\u5b57\u7b26\u4e32\u65f6，\u5b83\u8fd8\u4f1a\u5c1d\u8bd5\u67e5\u770b\u6570\u636e\u662f\u5426\u5339\u914d\u65e5\u671f\u6a21\u5f0f\u3002\u5982\u679c\u9700\u8981，\u4e5f\u53ef\u4ee5\u4e3a\u6570\u5b57\u542f\u7528\u57fa\u4e8e\u6a21\u5f0f\u7684\u5b57\u7b26\u4e32\u68c0\u6d4b\u3002<\/p>\n
text \u5b57\u6bb5\u5177\u6709\u4e00\u4e2a\u5e26\u6709 keyword \u6761\u76ee\u7684 fields \u5b9a\u4e49\u3002\u8fd9\u79f0\u4e3a\u5b50\u5b57\u6bb5，\u662f\u4e00\u79cd\u5728\u9002\u5f53\u60c5\u51b5\u4e0b\u53ef\u7528\u7684\u66ff\u4ee3\u7c7b\u578b\u6216\u8f85\u52a9\u7c7b\u578b\u3002\u5728 Elasticsearch \u4e2d，\u52a8\u6001\u7c7b\u578b\u7684 text \u5b57\u6bb5\u4f1a\u88ab\u8d4b\u4e88\u4e00\u4e2a keyword \u5b50\u5b57\u6bb5\u3002\u4f60\u5df2\u7ecf\u4f7f\u7528 category.keyword \u5b50\u5b57\u6bb5\u6765\u5bf9\u7ed9\u5b9a\u7c7b\u522b\u6267\u884c\u7cbe\u786e\u641c\u7d22\u3002\u4e3a\u4e86\u907f\u514d\u6dfb\u52a0\u5b50\u5b57\u6bb5，\u53ef\u4ee5\u63d0\u4f9b\u4e00\u4e2a\u663e\u5f0f\u7684 text \u6216 keyword \u6620\u5c04，\u7136\u540e\u8be5\u7c7b\u578b\u5c06\u6210\u4e3a\u4e3b\u8981\u4e14\u552f\u4e00\u7684\u7c7b\u578b\u3002<\/p>\n<\/p>\n
\u8ba9\u6211\u4eec\u5411\u7d22\u5f15\u6dfb\u52a0\u4e00\u4e2a\u65b0\u5b57\u6bb5，\u7528\u4e8e\u5b58\u50a8\u6bcf\u4e2a\u6587\u6863\u7684 embedding\u3002<\/p>\n
\u663e\u5f0f\u6620\u5c04\u7684\u7ed3\u6784\u4e0e Elasticsearch \u5ba2\u6237\u7aef\u7684 get_mapping() \u65b9\u6cd5\u8fd4\u56de\u7684\u54cd\u5e94\u4e2d\u7684 mappings \u952e\u76f8\u5339\u914d\u3002\u53ea\u9700\u8981\u63d0\u4f9b\u9700\u8981\u663e\u5f0f\u6307\u5b9a\u7c7b\u578b\u7684\u5b57\u6bb5，\u56e0\u4e3a\u6620\u5c04\u4e2d\u672a\u5305\u542b\u7684\u4efb\u4f55\u5b57\u6bb5\u4ecd\u4f1a\u50cf\u4e4b\u524d\u4e00\u6837\u88ab\u52a8\u6001\u8bbe\u7f6e\u7c7b\u578b\u3002<\/p>\n
\u4e0b\u9762\u4f60\u53ef\u4ee5\u770b\u5230 Search \u7c7b\u4e2d create_index() \u65b9\u6cd5\u7684\u65b0\u7248\u672c，\u5b83\u6dfb\u52a0\u4e86\u4e00\u4e2a\u540d\u4e3a embedding \u7684\u663e\u5f0f\u7c7b\u578b\u5b57\u6bb5\u3002\u5728 search.py \u4e2d\u66ff\u6362\u6b64\u65b9\u6cd5：<\/p>\n
class Search:
\n # …<\/p>\n
def create_index(self):
\n self.es.indices.delete(index='my_documents', ignore_unavailable=True)
\n self.es.indices.create(index='my_documents', mappings={
\n 'properties': {
\n 'embedding': {
\n 'type': 'dense_vector',
\n }
\n }
\n }) <\/p>\n
\u6b63\u5982\u4f60\u6240\u770b\u5230\u7684，embedding \u5b57\u6bb5\u88ab\u8d4b\u4e88\u4e86 dense_vector \u7c7b\u578b，\u8fd9\u662f\u5b58\u50a8 embeddings \u65f6\u5408\u9002\u7684\u7c7b\u578b\u3002\u7a0d\u540e\u4f60\u5c06\u5b66\u4e60\u53e6\u4e00\u79cd\u5411\u91cf\u7c7b\u578b sparse_vector，\u5b83\u5728\u5176\u4ed6\u7c7b\u578b\u7684\u8bed\u4e49\u641c\u7d22\u5e94\u7528\u4e2d\u5f88\u6709\u7528\u3002<\/p>\n
dense_vector \u7c7b\u578b\u63a5\u53d7\u51e0\u4e2a\u53c2\u6570，\u6240\u6709\u53c2\u6570\u90fd\u662f\u53ef\u9009\u7684\u3002<\/p>\n
\u5728\u4e0a\u4e00\u8282\u4e2d，\u4f60\u5df2\u7ecf\u5b66\u4e60\u4e86\u5982\u4f55\u4f7f\u7528 SentenceTransformers \u6846\u67b6\u548c all-MiniLM-L6-v2 \u6a21\u578b\u751f\u6210 embeddings\u3002\u73b0\u5728\u662f\u65f6\u5019\u5c06\u8be5\u6a21\u578b\u96c6\u6210\u5230\u5e94\u7528\u7a0b\u5e8f\u4e2d\u4e86\u3002<\/p>\n
\u9996\u5148，\u53ef\u4ee5\u5728 Search \u7c7b\u7684\u6784\u9020\u51fd\u6570\u4e2d\u5b9e\u4f8b\u5316\u8be5\u6a21\u578b：<\/p>\n
# …
\nfrom sentence_transformers import SentenceTransformer<\/p>\n
# …<\/p>\n
class Search:
\n def __init__(self):
\n self.model = SentenceTransformer('all-MiniLM-L6-v2')
\n self.es = Elasticsearch(cloud_id=os.environ['ELASTIC_CLOUD_ID'],
\n api_key=os.environ['ELASTIC_API_KEY'])
\n client_info = self.es.info()
\n print('Connected to Elasticsearch!')
\n pprint(client_info.body)<\/p>\n
# … <\/p>\n
\u6b63\u5982\u4f60\u5728\u672c\u6559\u7a0b\u7684\u5168\u6587\u641c\u7d22\u90e8\u5206\u4e2d\u8bb0\u5f97\u7684\u90a3\u6837，Search \u7c7b\u5177\u6709 insert_document() \u548c insert_documents() \u65b9\u6cd5，\u5206\u522b\u7528\u4e8e\u5411\u7d22\u5f15\u4e2d\u63d2\u5165\u5355\u4e2a\u6587\u6863\u548c\u591a\u4e2a\u6587\u6863\u3002\u8fd9\u4e24\u4e2a\u65b9\u6cd5\u73b0\u5728\u9700\u8981\u751f\u6210\u4e0e\u6bcf\u4e2a\u6587\u6863\u5bf9\u5e94\u7684 embeddings\u3002<\/p>\n
\u4e0b\u9762\u7684\u4ee3\u7801\u5757\u5c55\u793a\u4e86\u8fd9\u4e24\u4e2a\u65b9\u6cd5\u7684\u65b0\u7248\u672c，\u4ee5\u53ca\u4e00\u4e2a\u65b0\u7684 get_embedding() \u8f85\u52a9\u65b9\u6cd5，\u8be5\u65b9\u6cd5\u8fd4\u56de\u4e00\u4e2a embedding\u3002<\/p>\n
class Search:
\n # …<\/p>\n
def get_embedding(self, text):
\n return self.model.encode(text)<\/p>\n
def insert_document(self, document):
\n return self.es.index(index='my_documents', document={
\n **document,
\n 'embedding': self.get_embedding(document['summary']),
\n })<\/p>\n
def insert_documents(self, documents):
\n operations = []
\n for document in documents:
\n operations.append({'index': {'_index': 'my_documents'}})
\n operations.append({
\n **document,
\n 'embedding': self.get_embedding(document['summary']),
\n })
\n return self.es.bulk(operations=operations) <\/p>\n
\u4fee\u6539\u540e\u7684\u65b9\u6cd5\u5411\u8981\u63d2\u5165\u7684\u6587\u6863\u6dfb\u52a0\u4e86\u65b0\u7684 embedding \u5b57\u6bb5\u3002\u8be5 embedding \u662f\u6839\u636e\u6bcf\u4e2a\u6587\u6863\u7684 summary \u5b57\u6bb5\u751f\u6210\u7684\u3002\u901a\u5e38\u60c5\u51b5\u4e0b，embeddings \u662f\u4ece\u53e5\u5b50\u6216\u77ed\u6bb5\u843d\u4e2d\u751f\u6210\u7684，\u56e0\u6b64\u5728\u8fd9\u79cd\u60c5\u51b5\u4e0b，summary \u662f\u4e00\u4e2a\u7406\u60f3\u7684\u4f7f\u7528\u5b57\u6bb5\u3002\u5176\u4ed6\u53ef\u9009\u9879\u5305\u62ec name \u5b57\u6bb5，\u5b83\u5305\u542b\u6587\u6863\u7684\u6807\u9898，\u6216\u8005\u6587\u6863\u6b63\u6587\u4e2d\u7684\u524d\u51e0\u53e5\u8bdd\u3002<\/p>\n
\u5b8c\u6210\u8fd9\u4e9b\u66f4\u6539\u540e，\u53ef\u4ee5\u91cd\u65b0\u6784\u5efa\u7d22\u5f15，\u4f7f\u5176\u4e3a\u6bcf\u4e2a\u6587\u6863\u5b58\u50a8\u4e00\u4e2a embedding\u3002\u8981\u91cd\u65b0\u6784\u5efa\u7d22\u5f15，\u8bf7\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4：<\/p>\n
flask run <\/p>\n
\u5f53\u6211\u4eec\u91cd\u65b0\u8fd0\u884c\u4e0a\u9762\u7684\u547d\u4ee4\u540e，\u8fd9\u6b21 flask \u7684\u542f\u52a8\u6bd4\u8f83\u6162\u3002\u8fd9\u662f\u56e0\u4e3a\u5b83\u9700\u8981\u4e0b\u8f7d\u6a21\u578b all-MiniLM-L6-v2\u3002<\/p>\n
flask reindex
\n(.venv) $ flask reindex
\nWarning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
\nLoading weights: 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 103\/103 [00:00<00:00, 30637.07it\/s]
\nConnected to Elasticsearch!
\n{'cluster_name': 'elasticsearch',
\n 'cluster_uuid': 'G5xn8TT3StGeUFd9WJMy5A',
\n 'name': 'liuxgn.local',
\n 'tagline': 'You Know, for Search',
\n 'version': {'build_date': '2026-04-30T15:05:34.751113474Z',
\n 'build_flavor': 'default',
\n 'build_hash': '2e8528e92361c3399724226deaf2b46f933e925b',
\n 'build_snapshot': False,
\n 'build_type': 'tar',
\n 'lucene_version': '10.4.0',
\n 'minimum_index_compatibility_version': '8.0.0',
\n 'minimum_wire_compatibility_version': '8.19.0',
\n 'number': '9.4.0'}}
\nreindexing …
\nCreating the index …
\nIndex with 15 documents created in 205 milliseconds. <\/p>\n
\u5982\u679c\u4f60\u6709 Kibana \u5728\u8fd0\u884c，\u90a3\u4e48\u4f60\u53ef\u4ee5\u68c0\u67e5\u6700\u65b0\u7684 mapping：<\/p>\n
GET my_documents\/_mapping <\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802010102-6a6e96ce32f69.png\")
<\/p>\n
\u5982\u679c\u4f60\u9700\u8981\u63d0\u9192\u4e00\u4e0b，flask reindex \u547d\u4ee4\u662f\u5728 app.py \u6587\u4ef6\u4e2d\u7684 reindex() \u51fd\u6570\u91cc\u5b9e\u73b0\u7684\u3002\u5b83\u8c03\u7528 Search \u7c7b\u7684 reindex() \u65b9\u6cd5，\u800c\u8be5\u65b9\u6cd5\u4f9d\u6b21\u8c03\u7528 create_index()，\u7136\u540e\u5c06 data.json \u6587\u4ef6\u4e2d\u7684\u6240\u6709\u6570\u636e\u4f20\u9012\u7ed9 insert_documents()\u3002<\/p>\n<\/p>\n
k \u8fd1\u90bb（kNN）\u7b97\u6cd5\u4f1a\u5bf9 dense_vector \u7c7b\u578b\u7684\u5b57\u6bb5\u6267\u884c\u76f8\u4f3c\u5ea6\u641c\u7d22\u3002\u8fd9\u79cd\u641c\u7d22\u66f4\u51c6\u786e\u5730\u79f0\u4e3a\u201c\u8fd1\u4f3c kNN\u201d，\u5b83\u63a5\u53d7\u4e00\u4e2a\u5411\u91cf\u6216 embedding \u4f5c\u4e3a\u641c\u7d22\u8bcd，\u5e76\u5728\u7d22\u5f15\u4e2d\u67e5\u627e\u63a5\u8fd1\u8be5\u5411\u91cf\u7684\u6761\u76ee\u3002<\/p>\n
\u5728\u672c\u8282\u4e2d，\u4f60\u5c06\u5b66\u4e60\u5982\u4f55\u4f7f\u7528\u4e0a\u4e00\u8282\u4e2d\u521b\u5efa\u7684\u6587\u6863 embeddings \u6267\u884c kNN \u641c\u7d22\u3002<\/p>\n<\/p>\n
\u5728\u672c\u6559\u7a0b\u7684\u5168\u6587\u641c\u7d22\u90e8\u5206\u4e2d，\u4f60\u5df2\u7ecf\u5b66\u4e60\u4e86\u4f20\u9012\u7ed9 Elasticsearch \u5ba2\u6237\u7aef search() \u65b9\u6cd5\u7684 query \u9009\u9879\u3002\u5728\u641c\u7d22\u5411\u91cf\u65f6，\u5219\u4f7f\u7528 knn \u9009\u9879\u3002<\/p>\n
\u4e0b\u9762\u4f60\u53ef\u4ee5\u770b\u5230 app.py \u4e2d handle_search() \u51fd\u6570\u7684\u65b0\u7248\u672c，\u8be5\u51fd\u6570\u4f1a\u9488\u5bf9\u7528\u6237\u5728\u641c\u7d22\u8868\u5355\u4e2d\u8f93\u5165\u7684\u67e5\u8be2\u8fd0\u884c kNN \u641c\u7d22\u3002<\/p>\n
@app.post('\/')
\ndef handle_search():
\n query = request.form.get('query', '')
\n filters, parsed_query = extract_filters(query)
\n from_ = request.form.get('from_', type=int, default=0)<\/p>\n
results = es.search(
\n knn={
\n 'field': 'embedding',
\n 'query_vector': es.get_embedding(parsed_query),
\n 'num_candidates': 50,
\n 'k': 10,
\n },
\n size=5,
\n from_=from_
\n )
\n return render_template('index.html', results=results['hits']['hits'],
\n query=query, from_=from_,
\n total=results['hits']['total']['value']) <\/p>\n
\u5728\u8fd9\u4e2a\u7248\u672c\u7684\u51fd\u6570\u4e2d，query \u9009\u9879\u88ab\u66ff\u6362\u4e3a knn\u3002\u7528\u4e8e\u5206\u9875\u7684 size \u548c from_ \u9009\u9879\u4fdd\u6301\u4e0d\u53d8，\u51fd\u6570\u4e2d\u7684\u5176\u4ed6\u6240\u6709\u5185\u5bb9\u4ee5\u53ca index.html \u6a21\u677f\u4e5f\u4e0e\u4e4b\u524d\u76f8\u540c\u3002<\/p>\n
knn \u641c\u7d22\u9009\u9879\u63a5\u53d7\u591a\u4e2a\u7528\u4e8e\u914d\u7f6e\u641c\u7d22\u7684\u53c2\u6570：<\/p>\n
\u4f7f\u7528\u4e0a\u8ff0\u4ee3\u7801\u4e2d\u7684\u8bbe\u7f6e，\u5c06\u8fd4\u56de 10 \u4e2a\u6700\u4f73\u5339\u914d\u7ed3\u679c\u3002<\/p>\n
\u6b22\u8fce\u4f60\u5c1d\u8bd5\u8fd9\u4e2a\u65b0\u7248\u672c\u7684\u5e94\u7528\u7a0b\u5e8f\u3002\u4ee5\u4e0b\u662f\u4e24\u4e2a\u5f88\u597d\u7684\u793a\u4f8b，\u53ef\u4ee5\u5e2e\u52a9\u4f60\u7406\u89e3\u8fd9\u79cd\u641c\u7d22\u7c7b\u578b\u7684\u5b9e\u7528\u6027：<\/p>\n
\u6b63\u5982\u672c\u6559\u7a0b\u5168\u6587\u641c\u7d22\u90e8\u5206\u4e2d\u5b9a\u4e49\u7684\u90a3\u6837，\u641c\u7d22\u67e5\u8be2\u5141\u8bb8\u7528\u6237\u4f7f\u7528 category: \u8bed\u6cd5\u5728\u641c\u7d22\u6587\u672c\u7684\u4efb\u610f\u4f4d\u7f6e\u8bf7\u6c42\u4f7f\u7528\u7279\u5b9a\u7c7b\u522b\u3002app.py \u4e2d\u7684 extract_filters() \u51fd\u6570\u8d1f\u8d23\u67e5\u627e\u5e76\u4ece\u641c\u7d22\u67e5\u8be2\u4e2d\u5206\u79bb\u8fd9\u4e9b\u8fc7\u6ee4\u8868\u8fbe\u5f0f\u3002\u5728\u4e0a\u4e00\u8282\u4e2d\u7684 handle_search() \u51fd\u6570\u7248\u672c\u91cc，filters \u53d8\u91cf\u6ca1\u6709\u88ab\u4f7f\u7528，\u56e0\u6b64 category \u8fc7\u6ee4\u5668\u4f1a\u88ab\u5ffd\u7565\u3002<\/p>\n
\u5e78\u8fd0\u7684\u662f，knn \u9009\u9879\u4e5f\u652f\u6301\u8fc7\u6ee4\u529f\u80fd\u3002filter \u9009\u9879\u5b9e\u9645\u4e0a\u63a5\u53d7\u76f8\u540c\u7c7b\u578b\u7684\u8fc7\u6ee4\u5668，\u56e0\u6b64\u53ef\u4ee5\u5c06\u8fc7\u6ee4\u5668\u76f4\u63a5\u63d2\u5165 knn \u67e5\u8be2\u4e2d，\u5b8c\u5168\u6309\u7167 extract_filters() \u51fd\u6570\u8fd4\u56de\u7684\u5f62\u5f0f\u4f7f\u7528：<\/p>\n
@app.post('\/')
\ndef handle_search():
\n query = request.form.get('query', '')
\n filters, parsed_query = extract_filters(query)
\n from_ = request.form.get('from_', type=int, default=0)<\/p>\n
results = es.search(
\n knn={
\n 'field': 'embedding',
\n 'query_vector': es.get_embedding(parsed_query),
\n 'k': 10,
\n 'num_candidates': 50,
\n **filters,
\n },
\n size=5,
\n from_=from_
\n )
\n return render_template('index.html', results=results['hits']['hits'],
\n query=query, from_=from_,
\n total=results['hits']['total']['value']) <\/p>\n
\u805a\u5408\u4e5f\u53ef\u4ee5\u5f88\u597d\u5730\u7528\u4e8e kNN \u67e5\u8be2，\u56e0\u6b64\u4e5f\u53ef\u4ee5\u5c06\u5b83\u4eec\u91cd\u65b0\u6dfb\u52a0\u56de\u6765：<\/p>\n
@app.post('\/')
\ndef handle_search():
\n query = request.form.get('query', '')
\n filters, parsed_query = extract_filters(query)
\n from_ = request.form.get('from_', type=int, default=0)<\/p>\n
results = es.search(
\n knn={
\n 'field': 'embedding',
\n 'query_vector': es.get_embedding(parsed_query),
\n 'k': 10,
\n 'num_candidates': 50,
\n **filters,
\n },
\n aggs={
\n 'category-agg': {
\n 'terms': {
\n 'field': 'category.keyword',
\n }
\n },
\n 'year-agg': {
\n 'date_histogram': {
\n 'field': 'updated_at',
\n 'calendar_interval': 'year',
\n 'format': 'yyyy',
\n },
\n },
\n },
\n size=5,
\n from_=from_
\n )
\n aggs = {
\n 'Category': {
\n bucket['key']: bucket['doc_count']
\n for bucket in results['aggregations']['category-agg']['buckets']
\n },
\n 'Year': {
\n bucket['key_as_string']: bucket['doc_count']
\n for bucket in results['aggregations']['year-agg']['buckets']
\n if bucket['doc_count'] > 0
\n },
\n }
\n return render_template('index.html', results=results['hits']['hits'],
\n query=query, from_=from_,
\n total=results['hits']['total']['value'], aggs=aggs) <\/p>\n
\u8fd9\u4e2a\u7248\u672c\u7684 handle_search() \u51fd\u6570\u5177\u6709\u4e0e\u5168\u6587\u641c\u7d22\u7248\u672c\u76f8\u540c\u7684\u529f\u80fd，\u53ea\u4e0d\u8fc7\u4f7f\u7528\u5411\u91cf\u641c\u7d22\u4ee3\u66ff\u4e86\u57fa\u4e8e\u5173\u952e\u8bcd\u7684\u641c\u7d22\u3002<\/p>\n
\u5728\u4e0b\u4e00\u8282\u4e2d，\u4f60\u5c06\u5b66\u4e60\u5982\u4f55\u7ed3\u5408\u8fd9\u4e24\u79cd\u4e0d\u540c\u641c\u7d22\u65b9\u6cd5\u7684\u7ed3\u679c\u3002<\/p>\n<\/p>\n
\u73b0\u5728\u4f60\u5df2\u7ecf\u4e86\u89e3\u4e86\u4e24\u79cd\u4e0d\u540c\u7684\u6587\u6863\u96c6\u5408\u641c\u7d22\u65b9\u6cd5，\u6bcf\u79cd\u65b9\u6cd5\u90fd\u6709\u5176\u72ec\u7279\u7684\u4f18\u52bf\u3002\u5982\u679c\u5176\u4e2d\u4e00\u79cd\u65b9\u6cd5\u7b26\u5408\u4f60\u7684\u9700\u6c42，\u90a3\u4e48\u4f60\u4e0d\u9700\u8981\u5176\u4ed6\u65b9\u6cd5，\u4f46\u5728\u5f88\u591a\u60c5\u51b5\u4e0b，\u6bcf\u79cd\u641c\u7d22\u65b9\u6cd5\u90fd\u4f1a\u8fd4\u56de\u53e6\u4e00\u79cd\u65b9\u6cd5\u53ef\u80fd\u9057\u6f0f\u7684\u6709\u4ef7\u503c\u7ed3\u679c，\u56e0\u6b64\u6700\u4f73\u9009\u62e9\u662f\u63d0\u4f9b\u4e00\u4e2a\u7ec4\u5408\u540e\u7684\u7ed3\u679c\u96c6\u3002<\/p>\n
\u9488\u5bf9\u8fd9\u4e9b\u60c5\u51b5，Elasticsearch \u63d0\u4f9b\u4e86 Reciprocal Rank Fusion，\u8fd9\u662f\u4e00\u79cd\u5c06\u4e24\u4e2a\u6216\u591a\u4e2a\u5217\u8868\u4e2d\u7684\u7ed3\u679c\u5408\u5e76\u4e3a\u4e00\u4e2a\u5217\u8868\u7684\u7b97\u6cd5\u3002<\/p>\n<\/p>\n
Elasticsearch \u5c06 RRF \u7b97\u6cd5\u96c6\u6210\u5230\u4e86\u641c\u7d22\u67e5\u8be2\u4e2d\u3002\u8bf7\u8003\u8651\u4e0b\u9762\u7684\u793a\u4f8b，\u5176\u4e2d\u5305\u542b query \u548c knn \u90e8\u5206，\u5206\u522b\u7528\u4e8e\u8bf7\u6c42\u5168\u6587\u641c\u7d22\u548c\u5411\u91cf\u641c\u7d22，\u4ee5\u53ca\u4e00\u4e2a rrf \u90e8\u5206，\u7528\u4e8e\u5c06\u5b83\u4eec\u5408\u5e76\u4e3a\u4e00\u4e2a\u7ed3\u679c\u5217\u8868\u3002<\/p>\n
self.es.search(
\n query={
\n # full-text search query here
\n },
\n knn={
\n # vector search query here
\n },
\n rank={
\n "rrf": {}
\n }
\n) <\/p>\n
\u867d\u7136 RRF \u5728\u4e0d\u8fdb\u884c\u4efb\u4f55\u914d\u7f6e\u7684\u60c5\u51b5\u4e0b\u5bf9\u8f83\u77ed\u7684\u7ed3\u679c\u5217\u8868\u6548\u679c\u76f8\u5f53\u597d，\u4f46\u6709\u4e00\u4e9b\u53c2\u6570\u53ef\u4ee5\u8c03\u6574，\u4ee5\u63d0\u4f9b\u6700\u4f73\u7ed3\u679c\u3002\u8bf7\u67e5\u9605\u6587\u6863\u4ee5\u8be6\u7ec6\u4e86\u89e3\u8fd9\u4e9b\u53c2\u6570\u3002<\/p>\n<\/p>\n
\u8981\u542f\u7528\u4e00\u4e2a\u540c\u65f6\u8fd4\u56de\u5168\u6587\u641c\u7d22\u548c\u5411\u91cf\u641c\u7d22\u65b9\u6cd5\u7ed3\u679c\u7684\u7ec4\u5408\u641c\u7d22，\u9700\u8981\u6062\u590d\u4e4b\u524d\u5728 handle_search() \u51fd\u6570\u4e2d\u4f7f\u7528\u7684\u5168\u6587\u641c\u7d22\u903b\u8f91\u3002\u4e3a\u4e86\u5b9e\u73b0\u6df7\u5408\u641c\u7d22\u7b56\u7565，search() \u65b9\u6cd5\u5fc5\u987b\u540c\u65f6\u63a5\u6536 query \u548c knn \u53c2\u6570，\u6bcf\u4e2a\u53c2\u6570\u5206\u522b\u8bf7\u6c42\u4e00\u4e2a\u72ec\u7acb\u7684\u67e5\u8be2\u3002\u8fd8\u9700\u8981\u6dfb\u52a0\u4e0a\u9762\u6240\u793a\u7684 rank \u90e8\u5206，\u4ee5\u5c06\u7ed3\u679c\u5408\u5e76\u4e3a\u4e00\u4e2a\u6392\u5e8f\u5217\u8868\u3002<\/p>\n
\u4e0b\u9762\u662f\u5b9e\u73b0\u6df7\u5408\u641c\u7d22\u7b56\u7565\u7684 handle_search() \u51fd\u6570\u7248\u672c：<\/p>\n
@app.post('\/')
\ndef handle_search():
\n query = request.form.get('query', '')
\n filters, parsed_query = extract_filters(query)
\n from_ = request.form.get('from_', type=int, default=0)<\/p>\n
if parsed_query:
\n search_query = {
\n 'must': {
\n 'multi_match': {
\n 'query': parsed_query,
\n 'fields': ['name', 'summary', 'content'],
\n }
\n }
\n }
\n else:
\n search_query = {
\n 'must': {
\n 'match_all': {}
\n }
\n }<\/p>\n
results = es.search(
\n query={
\n 'bool': {
\n **search_query,
\n **filters
\n }
\n },
\n knn={
\n 'field': 'embedding',
\n 'query_vector': es.get_embedding(parsed_query),
\n 'k': 10,
\n 'num_candidates': 50,
\n **filters,
\n },
\n rank={
\n 'rrf': {}
\n },
\n aggs={
\n 'category-agg': {
\n 'terms': {
\n 'field': 'category.keyword',
\n }
\n },
\n 'year-agg': {
\n 'date_histogram': {
\n 'field': 'updated_at',
\n 'calendar_interval': 'year',
\n 'format': 'yyyy',
\n },
\n },
\n },
\n size=5,
\n from_=from_,
\n )
\n aggs = {
\n 'Category': {
\n bucket['key']: bucket['doc_count']
\n for bucket in results['aggregations']['category-agg']['buckets']
\n },
\n 'Year': {
\n bucket['key_as_string']: bucket['doc_count']
\n for bucket in results['aggregations']['year-agg']['buckets']
\n if bucket['doc_count'] > 0
\n },
\n }
\n return render_template('index.html', results=results['hits']['hits'],
\n query=query, from_=from_,
\n total=results['hits']['total']['value'], aggs=aggs) <\/p>\n
\u4f7f\u7528\u8fd9\u4e2a\u7248\u672c\u540e，\u6bcf\u79cd\u641c\u7d22\u65b9\u6cd5\u7684\u6700\u4f73\u7ed3\u679c\u90fd\u4f1a\u88ab\u5408\u5e76\u3002\u70b9\u51fb\u6b64\u5904\u67e5\u770b\u5305\u542b\u8fd9\u4e9b\u66f4\u6539\u7684\u5b8c\u6574\u5e94\u7528\u7a0b\u5e8f\u3002<\/p>\n
\u6b63\u5bf9\u81ea\u7b7e\u540d\u7684 Elasticsearch \u96c6\u7fa4，\u8bf7\u53c2\u8003\u4ee3\u7801\u00a0GitHub – liu-xiao-guo\/search-tutorial-v2 \u00b7 GitHub<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802010103-6a6e96cfa661f.png\")
<\/p>\n","protected":false},"excerpt":{"rendered":"
\u8fd9\u7bc7\u6587\u7ae0\u662f\u7ee7\u4e4b\u524d\u6587\u7ae0 \u201cElasticsearch：\u641c\u7d22\u6559\u7a0b – \u5168\u6587\u641c\u7d22（\u4e00）\u201d\u3002\u5728\u7ee7\u7eed\u8fd9\u7bc7\u6587\u7ae0\u4e4b\u524d，\u4f60\u5fc5\u987b\u4ed4\u7ec6\u9605\u8bfb\u4e4b\u524d\u7684\u90a3\u7bc7\u6587\u7ae0\u3002 Embedding \u7b80\u4ecb
\n\u5728 Machine Learning \u4e2d，embedding \u662f\u4e00\u4e2a vector（\u6570\u5b57\u6570\u7ec4），\u7528\u4e8e\u8868\u793a\u73b0\u5b9e\u4e16\u754c\u4e2d\u7684\u5bf9\u8c61，\u4f8b\u5982 words\u3001sentences\u3001images \u6216 videos<\/p>\n","protected":false},"author":2,"featured_media":89021,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[1299,3368,323,2215,100],"topic":[],"class_list":["post-89023","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-elasticsearch","tag-3368","tag-323","tag-2215","tag-100"],"yoast_head":"\n