大家好,我是微学AI,今天给大家带来自然语言处理实战项目4-文本相似度的搜索功能,搜索文本内容。文本相似度搜索是一种基于自然语言处理技术,用于搜索和匹配文本内容的方法。其主要目的是将用户输入的查询内容与已有的文本数据进行比较,并找到最相似的文本数据。
本文本以目标实现为导向,实战让大家跑通文本相似度的搜索功能。
1.首先加载与处理文件夹数据,本文以txt文件为例子,批量处理。
2.然后构建文件名和文件内容的索引文件。
3.在进行文档向量化与模型构建,生成向量模型
4.加载模型进行相似度的计算并返回。
5.后续可以新增文档到向量模型,可搜索到新加的文件
mport os
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
import pickle
from settings.app_config import project_config as fileconfig
import PyPDF2
import csv
index_file = 'index.pkl'
vectorizer_file = 'vectorizer.pkl'
index_file_path = 'file_path.pickle'
# 构建文件搜索引擎类
class FileSearchManage():
def __init__(self):
self.index_file_path = index_file_path
self.index_file = index_file
self.vectorizer_file = vectorizer_file
#读取文件
def read_files(self,folder_path):
files_data = {}
for file_name in os.listdir(folder_path):
if file_name.endswith(".txt"):
with open(os.path.join(folder_path, file_name), 'r', encoding='utf-8') as f:
files_data[file_name] = f.read().replace('\n', '')
return files_data
# 保存 csv 文件
def save_csv_files(self, folder_path,csv_path):
# 将信息写入csv文件
with open(csv_path, 'w', newline='', encoding='utf-8') as csvfile:
fieldnames = ['file_name', 'paragraph', 'content']
writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
writer.writeheader()
for file_name in os.listdir(folder_path):
if file_name.endswith(".pdf"):
path_name, paragraph = ocr_paragraph_pdf(os.path.join(folder_path, file_name))
for i, pa in enumerate(paragraph):
pa = pa.replace('\n', '').replace(' ', '')
if len(pa) > 4:
writer.writerow({'file_name': file_name,
'paragraph': i,
'content': pa})
# 读取pickle索引文件
def read_pickle(self,index_file_path):
with open(index_file_path, 'rb') as f:
index = pickle.load(f)
return index
#分词、去除停用词 处理
def preprocess_data(self,files_data):
processed_data = {}
for file_name, content in files_data.items():
# 在这里可以对文档内容进行预处理(例如:分词、去除停用词)
processed_data[file_name] = content
return processed_data
# 创建索引文件向量
def create_tfidf_index(self,processed_data):
vectorizer = TfidfVectorizer()
corpus = list(processed_data.values())
X = vectorizer.fit_transform(corpus)
return X, vectorizer
# 报错文件json
def save_file(self,index_file_path,files_data):
with open(index_file_path, 'wb') as f:
pickle.dump(files_data, f)
# 保存索引文件向量
def save_index(self,index, vectorizer, index_file, vectorizer_file):
with open(index_file, 'wb') as f:
pickle.dump(index, f)
with open(vectorizer_file, 'wb') as f:
pickle.dump(vectorizer, f)
# 加载索引文件向量
def load_index(self,index_file, vectorizer_file):
with open(index_file, 'rb') as f:
index = pickle.load(f)
with open(vectorizer_file, 'rb') as f:
vectorizer = pickle.load(f)
return index, vectorizer
def preprocess_query(self,query):
# 对查询进行预处理(例如:分词、去除停用词)
return query
# 文件查找函数
def search(self,query, index, vectorizer, files_data,num):
processed_query = self.preprocess_query(query)
query_vector = vectorizer.transform([processed_query])
cosine_similarities = cosine_similarity(index, query_vector)
top_file_indices = cosine_similarities.ravel().argsort()[-int(num):][::-1]
# print(top_file_indices)
results = []
for file_index in top_file_indices:
file_name = list(files_data.keys())[file_index]
file_content = files_data[file_name]
similarity = cosine_similarities[file_index][0]
results.append((file_name, file_content, similarity))
return sorted(results, key=lambda x: x[-1], reverse=True)
# return most_similar_file_name, most_similar_file_content
# 文件相似度计算
def search_similar_files(self,query,num):
files_data = self.read_pickle(self.index_file_path)
#processed_data = self.preprocess_data(files_data)
index, vectorizer = self.load_index(self.index_file, self.vectorizer_file)
result = self.search(query, index, vectorizer, files_data,num)
result = [x[0]+" "+str(x[2]) for x in result]
return result
# print('File content:', result_file_content)
# 获取文件内容
def get_content(self,filename):
files_data = self.read_pickle(self.index_file_path)
result = files_data[filename]
return result
# 新增新的索引文件
def add_new_file(self,file_path):
index,vectorizer = self.load_index(self.index_file, self.vectorizer_file)
files_data = self.read_pickle(self.index_file_path)
content =''
try:
if file_path.split('.')[-1]=='txt':
with open(file_path, 'r', encoding='utf-8') as f:
file_content = f.read().replace('\n', '')
files_data[file_path.split('/')[-1]] = file_content
if file_path.split('.')[-1] == 'pdf':
with open(file_path, 'rb') as f:
pdf_reader = PyPDF2.PdfFileReader(f)
# 获取PDF文件的页数
num_pages = pdf_reader.numPages
# 创建文本文件,并将PDF文件每一页的内容写入
for i in range(num_pages):
page = pdf_reader.getPage(i)
text = page.extractText().replace(' ', '')
content = content + text
file_content = content.replace('\n', '')
files_data[file_path.split('/')[-1]] = file_content
with open(self.index_file_path, 'wb') as f:
pickle.dump(files_data, f)
corpus = list(files_data.values())
X = vectorizer.fit_transform(corpus)
with open(self.index_file, 'wb') as f:
pickle.dump(X, f)
with open(self.vectorizer_file, 'wb') as f:
pickle.dump(vectorizer, f)
return 'successful'
except Exception as e:
print(e)
return 'fail'
2.构建文件夹导入函数
if __name__ == '__main__':
folder_path = '文件夹的地址' # 例如'E:/data'
def create_file(folder_path):
FileSearch = FileSearchManage()
files_data = FileSearch.read_files(folder_path)
processed_data = FileSearch.preprocess_data(files_data)
FileSearch.save_file(index_file_path,processed_data)
index, vectorizer = FileSearch.create_tfidf_index(processed_data)
FileSearch.save_index(index, vectorizer, index_file, vectorizer_file)
def file_search(query):
FileSearch = FileSearchManage()
files_data = FileSearch.read_pickle(index_file_path)
processed_data = FileSearch.preprocess_data(files_data)
index, vectorizer = FileSearch.load_index(index_file, vectorizer_file)
result_file_name = FileSearch.search(query, index, vectorizer, files_data,num=5)
print('File name:', result_file_name)
def file_add(folder_path):
FileSearch = FileSearchManage()
files_data = FileSearch.read_pdf_files(folder_path)
with open(index_file_path, 'wb') as f:
pickle.dump(files_data, f)
processed_data = FileSearch.preprocess_data(files_data)
index, vectorizer = FileSearch.create_tfidf_index(processed_data)
FileSearch.save_index(index, vectorizer, index_file, vectorizer_file)
index, vectorizer = FileSearch.load_index(index_file, vectorizer_file)
query = '*****'
result_file_name = FileSearch.search(query, index, vectorizer, files_data,num=5)
create_file(folder_path)
#query ='搜索语句'
#file_search(query)
#file_add(folder_path)
我们还可以根据自己的需求,添加新的文件,可以是txt,pdf的文件,pdf有的文件可以直接转文本,有的图片的需要OCR识别,这个可以接入OCR进行识别,使得系统更加的完善。欢迎大家进行关注与支持,有更多需求和合作的可以联系。