pytorch基于GloVe实现的词嵌入

news2025/2/4 22:08:11

PyTorch 实现 GloVe（Global Vectors for Word Representation） 的完整代码，使用 中文语料 进行训练，包括 共现矩阵构建、模型定义、训练和测试。

1. GloVe 介绍

基于词的共现信息（不像 Word2Vec 使用滑动窗口预测）
适合较大规模的数据（比 Word2Vec 更稳定）
学习出的词向量能捕捉语义信息（如类比关系）

import torch
import torch.nn as nn
import torch.optim as optim
import numpy as np
import jieba
from collections import Counter
from scipy.sparse import coo_matrix

# ========== 1. 数据预处理 ==========
corpus = [
    "我们 喜欢 深度 学习",
    "自然 语言 处理 是 有趣 的",
    "人工智能 改变 了 世界",
    "深度 学习 是 人工智能 的 重要 组成部分"
]

# 分词
tokenized_corpus = [list(jieba.cut(sentence)) for sentence in corpus]
vocab = set(word for sentence in tokenized_corpus for word in sentence)
word2idx = {word: idx for idx, word in enumerate(vocab)}
idx2word = {idx: word for word, idx in word2idx.items()}

# 计算共现矩阵
window_size = 2
co_occurrence = Counter()

for sentence in tokenized_corpus:
    indices = [word2idx[word] for word in sentence]
    for center_idx in range(len(indices)):
        center_word = indices[center_idx]
        for offset in range(-window_size, window_size + 1):
            context_idx = center_idx + offset
            if 0 <= context_idx < len(indices) and context_idx != center_idx:
                context_word = indices[context_idx]
                co_occurrence[(center_word, context_word)] += 1

# 转换为稀疏矩阵
rows, cols, values = zip(*[(c[0], c[1], v) for c, v in co_occurrence.items()])
X = coo_matrix((values, (rows, cols)), shape=(len(vocab), len(vocab)))


# ========== 2. 定义 GloVe 模型 ==========
class GloVe(nn.Module):
    def __init__(self, vocab_size, embedding_dim):
        super(GloVe, self).__init__()
        self.w_embeddings = nn.Embedding(vocab_size, embedding_dim)  # 中心词嵌入
        self.c_embeddings = nn.Embedding(vocab_size, embedding_dim)  # 上下文词嵌入
        self.w_bias = nn.Embedding(vocab_size, 1)  # 中心词偏置
        self.c_bias = nn.Embedding(vocab_size, 1)  # 上下文词偏置
        nn.init.xavier_uniform_(self.w_embeddings.weight)
        nn.init.xavier_uniform_(self.c_embeddings.weight)

    def forward(self, center, context, co_occur):
        w_emb = self.w_embeddings(center)
        c_emb = self.c_embeddings(context)
        w_bias = self.w_bias(center).squeeze()
        c_bias = self.c_bias(context).squeeze()
        dot_product = (w_emb * c_emb).sum(dim=1)
        loss = (dot_product + w_bias + c_bias - torch.log(co_occur + 1e-8)) ** 2
        return loss.mean()


# 初始化模型
embedding_dim = 10
model = GloVe(len(vocab), embedding_dim)

# ========== 3. 训练 GloVe ==========
criterion = nn.MSELoss()
optimizer = optim.Adam(model.parameters(), lr=0.01)
num_epochs = 100

# 转换数据
co_occurrence_tensor = torch.tensor(X.data, dtype=torch.float)
pairs = list(zip(X.row, X.col, co_occurrence_tensor))

for epoch in range(num_epochs):
    total_loss = 0
    np.random.shuffle(pairs)
    for center, context, co_occur in pairs:
        optimizer.zero_grad()
        loss = model(
            torch.tensor([center], dtype=torch.long),
            torch.tensor([context], dtype=torch.long),
            torch.tensor([co_occur], dtype=torch.float)  # 修正数据类型
        )
        loss.backward()
        optimizer.step()
        total_loss += loss.item()

    if (epoch + 1) % 10 == 0:
        print(f"Epoch [{epoch + 1}/{num_epochs}], Loss: {total_loss:.4f}")

# ========== 4. 获取词向量 ==========
word_vectors = model.w_embeddings.weight.data.numpy()


# ========== 5. 计算相似度 ==========
def most_similar(word, top_n=3):
    if word not in word2idx:
        return "单词不在词汇表中"

    word_vec = word_vectors[word2idx[word]].reshape(1, -1)
    similarities = np.dot(word_vectors, word_vec.T).squeeze()
    similar_idx = similarities.argsort()[::-1][1:top_n + 1]
    return [(idx2word[idx], similarities[idx]) for idx in similar_idx]


# 测试
test_words = ["深度", "学习", "人工智能"]
for word in test_words:
    print(f"【{word}】的相似单词:", most_similar(word))