有勇气的牛排博客
☰

sentence-transformers文本向量化|语义搜索|相似度计算

进群口令：博客

1 前言

1.1 什么是 sentence-transformers

sentence-transformers 是一个基于BERT和其变种模型的Python库，用于生成高质量句子向量(sentence embeddings)。这些向量可以用于以下场景：

  • 文本相似度计算
  • 语义搜索（Semantic Search）
  • 文本聚类
  • 多轮问答
  • 向量数据库索引（如FAISS、Milvus）

1.2 环境安装

python3.10

sentence-transformers==4.1.0

2 编码句子

加载一个预训练模型并编码句子。

# -*- coding: utf-8 -*-
from sentence_transformers import SentenceTransformer

model_path = "./tmp/model/all-MiniLM-L6-v2"

# 加载模型
model = SentenceTransformer(model_path)

# 编码句子为向量
sentences = ["韩立飞升仙界", "修仙是一种漫长的旅途"]
embeddings = model.encode(sentences)

for sentence, embedding in zip(sentences, embeddings):
    print(f"句子: {sentence}")
    print(f"向量: {embedding[:5]}...")  # 仅展示前5维

sentence-transformers加载一个预训练模型并编码句子|文本向量化

3 相似度计算

# -*- coding: utf-8 -*-
from sentence_transformers import util, SentenceTransformer

# 加载模型
model_path = "./tmp/model/all-MiniLM-L6-v2"
model = SentenceTransformer(model_path)

s1 = "韩立突破金丹期"
s2 = "韩立进入金丹阶段"

emb1 = model.encode(s1)
emb2 = model.encode(s2)

similarity = util.pytorch_cos_sim(emb1, emb2)
print("句子相似度：", similarity.item())

sentence-transformers相似度计算

4 语义搜索

# -*- coding: utf-8 -*-

from sentence_transformers import  util,SentenceTransformer

# 加载模型
model_path = "E:/blog_article/tmp/model/all-MiniLM-L6-v2"
model = SentenceTransformer(model_path)

# 构建语料库
corpus = [
    "韩立突破金丹期",
    "韩立在黑风山修炼",
    "韩立炼制了丹药",
    "王林在苍茫星修炼",
]
corpus_embedding = model.encode(corpus)

# 查询句子
query = "韩立结丹"
query_embedding = model.encode(query)

# 计算相似度得分
hits = util.semantic_search(query_embedding, corpus_embedding, top_k=3)

for hit in hits[0]:
    print(f"得分: {hit['score']:.4f}, 语句: {corpus[hit['corpus_id']]}")

sentence-transformers语义搜索

5 与 FAISS 搭配构建向量检索系统

faiss-cpu

# -*- coding: utf-8 -*-
import faiss
import numpy as np
from sentence_transformers import SentenceTransformer

# 加载模型
model_path = "E:/blog_article/tmp/model/all-MiniLM-L6-v2"
model = SentenceTransformer(model_path)

# 构建语料
corpus = ["修仙者", "炼丹炉", "灵根测试", "灵石交易"]
corpus_embeddings = model.encode(corpus)

# 转换为 float32
corpus_embeddings = np.array(corpus_embeddings).astype("float32")

# 创建 FAISS 索引
index = faiss.IndexFlatL2(corpus_embeddings.shape[1])
index.add(corpus_embeddings)

# 查询向量
query = "灵石购买"
query_embedding = model.encode([query]).astype("float32")

# 搜索
top_k = 2
distances, indices = index.search(query_embedding, top_k)

for idx, distance in zip(indices[0], distances[0]):
    print(f"距离: {distance:.4f}, 匹配: {corpus[idx]}")

sentence-transformers与 FAISS 搭配构建向量检索系统

评论区

×
×