搜索引擎简介
莫烦NLP学习总结系列:搜索引擎简介
扩展:多模态搜索:以文字搜图片等,阿里在视频搜索领域的探索:多模态搜索算法实践
搜索引擎搜索的过程
- 原始数据(数据库海量的数据,BigData)
- 批量召回(倒排索引)
- 粗排(TF-IDF:处理匹配排序)
- 精排
什么是倒排索引?
假设你开了家咨询公司,手上有100篇材料。这时有人来找你咨询NLP的问题,你会怎么在这100篇材料中找到合适的内容呢?
——用倒排索引的方式就是:我们在第一次拿到所有材料时,把它们通读一遍,然后构建关键词和文章的对应关系。当用户在搜索特定词的时候,比如“红”,就会直接返回“红”这个【关键词索引】下的文章列表。先构造索引的好处就是能够将这种索引,放在后续的搜索中复用,搜索也就变成了一种词语匹配加返回索引材料的过程。
搜索引擎常用技术方法:Term Frequency - Inverse Document Frequency (TF-IDF)
TF-IDF
中文名称:【词频-逆文本频率指数】,用来表达一篇文章的关键词信息。TF是一个局部信息(某篇文档中的词频TF),IDF是一个全局参数,用来判断一个词在所有文档中是不是垃圾信息。
1、将多篇文档的关键词数量建立成对应的多维词向量:[{单词:数量},… , {单词:数量}] , 并计算其TF-IDF
2、用户输入的搜索词组,计算搜索词组的TF-IDF
3、计算搜索词组和每篇文章的cosine距离
代码实现
思考过程
1、文档数据(手码,爬取,搜集),建立自己的数据库
手码测试数据
docs = [
"it is a good day, I like to stay here",
"I am happy to be here",
"I am bob",
"it is sunny today",
"I have a party today",
"it is a dog and that is a cat",
"there are dog and cat on the tree",
"I study hard this morning",
"today is a good day",
"tomorrow will be a good day",
"I like coffee, I like book and I like apple",
"I do not like it",
"I am kitty, I like bob",
"I do not care who like bob, but I like kitty",
"It is coffee time, bring your cup",
]
2、倒排索引建立(此处省略)
3、建立自身数据库文档的TF-IDF
对数据库文档进行切分,单词提取,建立字典1{单词:索引},字典2{索引:单词}
#将文档的单词转换成ID形式,这样便于后续通过ID进行统计。
docs_words = [d.replace(",", "").split(" ") for d in docs]
vocab = set(itertools.chain(*docs_words))
v2i = {v: i for i, v in enumerate(vocab)}
i2v = {i: v for v, i in v2i.items()}
获取所有文档的TF,此处为47×15的矩阵,分别代表47个单词,15个文档,矩阵中的值指每一篇文档中某单词出现的次数
def safe_log(x):
mask = x != 0
x[mask] = np.log(x[mask])
return x
tf_methods = {
"log": lambda x: np.log(1+x),
"augmented": lambda x: 0.5 + 0.5 * x / np.max(x, axis=1, keepdims=True),
"boolean": lambda x: np.minimum(x, 1),
"log_avg": lambda x: (1 + safe_log(x)) / (1 + safe_log(np.mean(x, axis=1, keepdims=True))),
}
def get_tf(method="log"):
# 词w 在 文档d 的TF=文档d 中 词w 总数
# term frequency: how frequent a word appears in a doc
# _tf是一个47×15的表格,47代表所有文本中出现的单词(不同)个数,15代表文档个数
_tf = np.zeros((len(vocab), len(docs)), dtype=np.float64) # [n_vocab, n_doc]
for i, d in enumerate(docs_words):
# counter代表了第i个文档种每一个单词出现的个数,用键值对的方式表示
counter = Counter(d)
for v in counter.keys():
_tf[v2i[v], i] = counter[v] / counter.most_common(1)[0][1]
# counter.most_common(1)[0][1]的意思是个数最多的单词的数量,为什么要除一个数可以理解为标准化
weighted_tf = tf_methods.get(method, None)
if weighted_tf is None:
raise ValueError
# 传入一个47×15的矩阵,47代表词数,15代表文档数
return weighted_tf(_tf)
获取所有文档的IDF,此处为一个47维的向量,代表47个词的区分度
idf_methods = {
#log: 词w 的IDF本质计算 IDF=log(所有文档数/所有文档中 词w 数)
"log": lambda x: 1 + np.log(len(docs) / (x+1)),
"prob": lambda x: np.maximum(0, np.log((len(docs) - x) / (x+1))),
"len_norm": lambda x: x / (np.sum(np.square(x))+1),
}
def get_idf(method="log"):
# inverse document frequency: low idf for a word appears in more docs, mean less important
# 统计所有单词47在所有文档中一共出现的次数,df是一个47维的向量
df = np.zeros((len(i2v), 1))
for i in range(len(i2v)):
d_count = 0
for d in docs_words:
# 如果第i个单词在文档d中出现,d_count计数
d_count += 1 if i2v[i] in d else 0
df[i, 0] = d_count
idf_fn = idf_methods.get(method, None)
if idf_fn is None:
raise ValueError
#传入47个单词的数目,返回经过idf处理后的47维向量
return idf_fn(df)
计算文档TF-IDF
tf = get_tf() # [n_vocab, n_doc]
idf = get_idf() # [n_vocab, 1]
tf_idf = tf * idf
4、输入搜索词组
q = "I get a coffee cup"
5、对搜索的词组计算TF-IDF
对输入词组进行词频(TF)的计算,结合文档数据库的IDF,计算出词组的TF-IDF
【注】此处代码为docs_score()的一部分,计算搜索语句与数据库文档相似的时候,顺便把搜索词组的TF-IDF给计算了
q_words = q.replace(",", "").split(" ")
# add unknown words
unknown_v = 0
for v in set(q_words):
if v not in v2i:
v2i[v] = len(v2i)
i2v[len(v2i)-1] = v
unknown_v += 1
if unknown_v > 0:
_idf = np.concatenate((idf, np.zeros((unknown_v, 1), dtype=np.float)), axis=0)
_tf_idf = np.concatenate((tf_idf, np.zeros((unknown_v, tf_idf.shape[1]), dtype=np.float)), axis=0)
else:
_idf, _tf_idf = idf, tf_idf
counter = Counter(q_words)
q_tf = np.zeros((len(_idf), 1), dtype=np.float) # [n_vocab, 1]
# 求q的tf
for v in counter.keys():
q_tf[v2i[v], 0] = counter[v]
q_vec = q_tf * _idf # [n_vocab, 1]
6、通过计算cosine距离判断相似度,列举出前n个最相似的关键文档
def cosine_similarity(q, _tf_idf):
unit_q = q / np.sqrt(np.sum(np.square(q), axis=0, keepdims=True))
unit_ds = _tf_idf / np.sqrt(np.sum(np.square(_tf_idf), axis=0, keepdims=True))
similarity = unit_ds.T.dot(unit_q).ravel()
return similarity
def docs_score(q, len_norm=False):
"""
获取多维向量q与tfidf的cosine距离,计为相关性分数
:param q:
:param len_norm:
:return:
"""
q_words = q.replace(",", "").split(" ")
# add unknown words
unknown_v = 0
for v in set(q_words):
if v not in v2i:
v2i[v] = len(v2i)
i2v[len(v2i)-1] = v
unknown_v += 1
if unknown_v > 0:
_idf = np.concatenate((idf, np.zeros((unknown_v, 1), dtype=np.float)), axis=0)
_tf_idf = np.concatenate((tf_idf, np.zeros((unknown_v, tf_idf.shape[1]), dtype=np.float)), axis=0)
else:
_idf, _tf_idf = idf, tf_idf
counter = Counter(q_words)
q_tf = np.zeros((len(_idf), 1), dtype=np.float) # [n_vocab, 1]
# 求q的tf
for v in counter.keys():
q_tf[v2i[v], 0] = counter[v]
q_vec = q_tf * _idf # [n_vocab, 1]
#余弦距离对比
q_scores = cosine_similarity(q_vec, _tf_idf)
if len_norm:
len_docs = [len(d) for d in docs_words]
q_scores = q_scores / np.array(len_docs)
return q_scores
全部代码
参考莫烦Github
我的代码
import numpy as np
from collections import Counter
import itertools
from visual import show_tfidf # this refers to visual.py in my [repo](https://github.com/MorvanZhou/NLP-Tutorials/)
docs = [
"it is a good day, I like to stay here",
"I am happy to be here",
"I am bob",
"it is sunny today",
"I have a party today",
"it is a dog and that is a cat",
"there are dog and cat on the tree",
"I study hard this morning",
"today is a good day",
"tomorrow will be a good day",
"I like coffee, I like book and I like apple",
"I do not like it",
"I am kitty, I like bob",
"I do not care who like bob, but I like kitty",
"It is coffee time, bring your cup",
]
#将文档的单词转换成ID形式,这样便于后续通过ID进行统计。
docs_words = [d.replace(",", "").split(" ") for d in docs]
vocab = set(itertools.chain(*docs_words))
v2i = {v: i for i, v in enumerate(vocab)}
i2v = {i: v for v, i in v2i.items()}
def safe_log(x):
mask = x != 0
x[mask] = np.log(x[mask])
return x
tf_methods = {
"log": lambda x: np.log(1+x),
"augmented": lambda x: 0.5 + 0.5 * x / np.max(x, axis=1, keepdims=True),
"boolean": lambda x: np.minimum(x, 1),
"log_avg": lambda x: (1 + safe_log(x)) / (1 + safe_log(np.mean(x, axis=1, keepdims=True))),
}
idf_methods = {
#log: 词w 的IDF本质计算 IDF=log(所有文档数/所有文档中 词w 数)
"log": lambda x: 1 + np.log(len(docs) / (x+1)),
"prob": lambda x: np.maximum(0, np.log((len(docs) - x) / (x+1))),
"len_norm": lambda x: x / (np.sum(np.square(x))+1),
}
def get_tf(method="log"):
# 词w 在 文档d 的TF=文档d 中 词w 总数
# term frequency: how frequent a word appears in a doc
# _tf是一个47×15的表格,47代表所有文本中出现的单词(不同)个数,15代表文档个数
_tf = np.zeros((len(vocab), len(docs)), dtype=np.float64) # [n_vocab, n_doc]
for i, d in enumerate(docs_words):
# counter代表了第i个文档种每一个单词出现的个数,用键值对的方式表示
counter = Counter(d)
for v in counter.keys():
_tf[v2i[v], i] = counter[v] / counter.most_common(1)[0][1]
# counter.most_common(1)[0][1]的意思是个数最多的单词的数量,为什么要除一个数可以理解为标准化
weighted_tf = tf_methods.get(method, None)
if weighted_tf is None:
raise ValueError
# 传入一个47×15的矩阵,47代表词数,15代表文档数
return weighted_tf(_tf)
def get_idf(method="log"):
# inverse document frequency: low idf for a word appears in more docs, mean less important
# 统计所有单词47在所有文档中一共出现的次数,df是一个47维的向量
df = np.zeros((len(i2v), 1))
for i in range(len(i2v)):
d_count = 0
for d in docs_words:
# 如果第i个单词在文档d中出现,d_count计数
d_count += 1 if i2v[i] in d else 0
df[i, 0] = d_count
idf_fn = idf_methods.get(method, None)
if idf_fn is None:
raise ValueError
#传入47个单词的数目,返回经过idf处理后的47维向量
return idf_fn(df)
def cosine_similarity(q, _tf_idf):
unit_q = q / np.sqrt(np.sum(np.square(q), axis=0, keepdims=True))
unit_ds = _tf_idf / np.sqrt(np.sum(np.square(_tf_idf), axis=0, keepdims=True))
similarity = unit_ds.T.dot(unit_q).ravel()
return similarity
def docs_score(q, len_norm=False):
"""
获取多维向量q与tfidf的cosine距离,计为相关性分数
:param q:
:param len_norm:
:return:
"""
q_words = q.replace(",", "").split(" ")
# add unknown words
unknown_v = 0
for v in set(q_words):
if v not in v2i:
v2i[v] = len(v2i)
i2v[len(v2i)-1] = v
unknown_v += 1
if unknown_v > 0:
_idf = np.concatenate((idf, np.zeros((unknown_v, 1), dtype=np.float)), axis=0)
_tf_idf = np.concatenate((tf_idf, np.zeros((unknown_v, tf_idf.shape[1]), dtype=np.float)), axis=0)
else:
_idf, _tf_idf = idf, tf_idf
counter = Counter(q_words)
q_tf = np.zeros((len(_idf), 1), dtype=np.float) # [n_vocab, 1]
# 求q的tf
for v in counter.keys():
q_tf[v2i[v], 0] = counter[v]
q_vec = q_tf * _idf # [n_vocab, 1]
#余弦距离对比
q_scores = cosine_similarity(q_vec, _tf_idf)
if len_norm:
len_docs = [len(d) for d in docs_words]
q_scores = q_scores / np.array(len_docs)
return q_scores
def get_keywords(n=2):
for c in range(3):
col = tf_idf[:, c]
#argsort返回的是数组值从小到大的索引值,[-n:]取最后的n个数
idx = np.argsort(col)[-n:]
print("doc{}, top{} keywords {}".format(c, n, [i2v[i] for i in idx]))
tf = get_tf() # [n_vocab, n_doc]
idf = get_idf() # [n_vocab, 1]
tf_idf = tf * idf # [n_vocab, n_doc]
print("tf shape(vecb in each docs): ", tf.shape)
print("\ntf samples:\n", tf[:2])
print("\nidf shape(vecb in all docs): ", idf.shape)
print("\nidf samples:\n", idf[:2])
print("\ntf_idf shape: ", tf_idf.shape)
print("\ntf_idf sample:\n", tf_idf[:2])
# test
# 返回前n个关键词
get_keywords()
q = "I get a coffee cup"
scores = docs_score(q)
d_ids = scores.argsort()[-3:][::-1]
print("\ntop 3 docs for '{}':\n{}".format(q, [docs[i] for i in d_ids]))
show_tfidf(tf_idf.T, [i2v[i] for i in range(tf_idf.shape[0])], "tfidf_matrix")
用sklearn实现
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
from visual import show_tfidf # this refers to visual.py in my [repo](https://github.com/MorvanZhou/NLP-Tutorials/)
docs = [
"it is a good day, I like to stay here",
"I am happy to be here",
"I am bob",
"it is sunny today",
"I have a party today",
"it is a dog and that is a cat",
"there are dog and cat on the tree",
"I study hard this morning",
"today is a good day",
"tomorrow will be a good day",
"I like coffee, I like book and I like apple",
"I do not like it",
"I am kitty, I like bob",
"I do not care who like bob, but I like kitty",
"It is coffee time, bring your cup",
]
vectorizer = TfidfVectorizer()
# tf_idf用csr_matrix来进行存储,即稀疏矩阵
# fit_transform:Learn vocabulary and idf, return document-term matrix.
tf_idf = vectorizer.fit_transform(docs)
#zip() 函数用于将可迭代的对象作为参数,将对象中对应的元素打包成一个个元组,然后返回由这些元组组成的列表。
# print("idf: ", zip(vectorizer.idf_, vectorizer.get_feature_names()))
print("idf: ", [(n, idf) for idf, n in zip(vectorizer.idf_, vectorizer.get_feature_names())])
print("v2i: ", vectorizer.vocabulary_)
q = "I get a coffee cup"
# transform: Transform documents to document-term matrix.
qtf_idf = vectorizer.transform([q])
res = cosine_similarity(tf_idf, qtf_idf)
res = res.ravel().argsort()[-3:]
print("\ntop 3 docs for '{}':\n{}".format(q, [docs[i] for i in res[::-1]]))
i2v = {i: v for v, i in vectorizer.vocabulary_.items()}
dense_tfidf = tf_idf.todense()
show_tfidf(dense_tfidf, [i2v[i] for i in range(dense_tfidf.shape[1])], "tfidf_sklearn_matrix")