pip install Levenshtein

import Levenshtein

def levenshtein_similarity(text1, text2):
    
    # 計算 Levenshtein 距離
    distance = Levenshtein.distance(text1, text2)
    # 計算相似度并保留一位小數(shù)
    max_len = max(len(text1), len(text2))
    similarity = round(1 - (distance / max_len), 1) if max_len > 0 else 1.0
    return similarity

# 示例用法
text1 = "kitten"
text2 = "sitting"
print(levenshtein_similarity(text1, text2))  # 輸出為 0.5

pip install sentence-transformers

from sentence_transformers import SentenceTransformer, util
import numpy as np

def semantic_similarity_sbert(str1, str2):
  
    model = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')
    
    # 編碼句子
    embeddings = model.encode([str1, str2])
    
    # 計算余弦相似度
    cosine_score = util.cos_sim(embeddings[0], embeddings[1])
    
    # 轉(zhuǎn)換為0-1范圍并保留一位小數(shù)
    similarity = float(np.clip(cosine_score.item(), 0.0, 1.0))
    return round(similarity, 1)

# 示例
print(semantic_similarity_sbert("我喜歡吃蘋果", "我愛吃水果"))  # 輸出類似: 0.8

pip install spacy

python -m spacy download en_core_web_md

python -m spacy download zh_core_web_lg

import spacy

def spacy_similarity(text1, text2):
    # 加載模型
    nlp = spacy.load('zh_core_web_lg')
    doc1 = nlp(text1)
    doc2 = nlp(text2)
    # 計算語義相似度并保留一位小數(shù)
    similarity = round(doc1.similarity(doc2), 1)
    return similarity

# 示例用法
text1 = "The cat sits on the mat."
text2 = "A cat is resting on a mat."
print(spacy_similarity(text1, text2))  # 輸出可能為 0.8

import spacy
import numpy as np

def semantic_similarity_spacy(str1, str2):
    """
    使用spaCy的詞向量計算語義相似度
    """
    nlp = spacy.load("zh_core_web_lg")
    
    doc1 = nlp(str1)
    doc2 = nlp(str2)
    
    # 計算余弦相似度
    similarity = doc1.similarity(doc2)
    
    # 確保在0-1范圍內(nèi)并保留一位小數(shù)
    return round(max(0.0, min(1.0, similarity)), 1)

# 示例
print(semantic_similarity_spacy("今天天氣很好", "明天的天氣不錯"))  # 輸出類似: 0.7

pip install tensorflow-hub tensorflow-text

import tensorflow_hub as hub
import tensorflow_text as text  # 必須導(dǎo)入
import numpy as np

def semantic_similarity_use(str1, str2):
    """

    """
    module = hub.load("https://tfhub.dev/google/universal-sentence-encoder-multilingual/3")
    
    # 編碼句子
    embeddings = module([str1, str2])
    
    # 計算余弦相似度
    similarity = np.inner(embeddings[0], embeddings[1]) / (
        np.linalg.norm(embeddings[0]) * np.linalg.norm(embeddings[1])
    )
    
    # 確保在0-1范圍內(nèi)并保留一位小數(shù)
    return round(max(0.0, min(1.0, similarity)), 1)

# 示例
print(semantic_similarity_use("這只貓很可愛", "那只貓咪非常漂亮"))  # 輸出類似: 0.8

pip install bert-serving-server bert-serving-client

from bert_serving.client import BertClient
import numpy as np

def semantic_similarity_bert(str1, str2):
    """

    """
    bc = BertClient()
    
    # 編碼句子
    embeddings = bc.encode([str1, str2])
    
    # 計算余弦相似度
    similarity = np.dot(embeddings[0], embeddings[1]) / (
        np.linalg.norm(embeddings[0]) * np.linalg.norm(embeddings[1])
    )
    
    # 確保在0-1范圍內(nèi)并保留一位小數(shù)
    return round(max(0.0, min(1.0, similarity)), 1)

# 示例
# print(semantic_similarity_bert("他正在跑步", "他在運動"))  # 輸出類似: 0.9

from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np

def semantic_similarity_tfidf(str1, str2):
    """
    使用TF-IDF和余弦相似度計算文本相似度
    基于詞頻統(tǒng)計，不是真正的語義相似度，但可以作為基線方法
    """
    vectorizer = TfidfVectorizer()
    tfidf_matrix = vectorizer.fit_transform([str1, str2])
    
    # 計算余弦相似度
    similarity = cosine_similarity(tfidf_matrix[0:1], tfidf_matrix[1:2])[0][0]
    
    # 確保在0-1范圍內(nèi)并保留一位小數(shù)
    return round(max(0.0, min(1.0, similarity)), 1)

# 示例
print(semantic_similarity_tfidf("深度學(xué)習(xí)", "神經(jīng)網(wǎng)絡(luò)"))  # 輸出類似: 0.6

import jieba
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
import Levenshtein


class SemanticSimilarity:
    def __init__(self):
        self.vectorizer = TfidfVectorizer()

    def tokenize(self, text):
        """中文分詞處理"""
        return ' '.join(jieba.lcut(text))

    def tfidf_cosine(self, str1, str2):
        """基于TF-IDF的余弦相似度(0-1)"""
        corpus = [self.tokenize(str1), self.tokenize(str2)]
        tfidf = self.vectorizer.fit_transform(corpus)
        return max(0.0, min(1.0, cosine_similarity(tfidf[0], tfidf[1])[0][0]))

    def edit_similarity(self, str1, str2):
        """標準化編輯距離相似度(0-1)"""
        distance = Levenshtein.distance(str1, str2)
        max_len = max(len(str1), len(str2))
        return max(0.0, min(1.0, 1 - (distance / max_len))) if max_len > 0 else 1.0

    def combined_similarity(self, str1, str2):
        """綜合語義相似度(保留1位小數(shù))"""
        tfidf_score = self.tfidf_cosine(str1, str2)
        edit_score = self.edit_similarity(str1, str2)
        return round(0.6 * tfidf_score + 0.4 * edit_score, 1)


if __name__ == '__main__':
    ss = SemanticSimilarity()
    str1 = "第十八條 【商標代理】外國人或者外國企業(yè)在中國申請商標注冊和辦理其他商標事宜的，應(yīng)當委托國家認可的具有商標代理資格的組織代理。"
    str2 = "第十八條 申請商標注冊或者辦理其他商標事宜，可以自行辦理，也可以委托依法設(shè)立的商標代理機構(gòu)辦理。"
    print(f"相似度: {ss.combined_similarity(str1, str2)}")