Python字符串模糊匹配库FuzzyWuzzy

在计算机科学中，字符串模糊匹配（fuzzy string matching）是一种近似地（而不是精确地）查找与模式匹配的字符串的技术。换句话说，字符串模糊匹配是一种搜索，即使用户拼错单词或只输入部分单词进行搜索，也能够找到匹配项。因此，它也被称为字符串近似匹配。

字符串模糊搜索可用于各种应用程序，例如：

• 拼写检查和拼写错误纠正程序。例如，用户在Google中键入“Missisaga”，将返回文字为“Showing results for mississauga”的点击列表。也就是说，即使用户输入缺少字符、有多余的字符或者有其他类型的拼写错误，搜索查询也会返回结果。
• 重复记录检查。例如，由于名称拼写不同(例如Abigail Martin和Abigail Martinez)在数据库中被多次列出。

这篇文章将解释字符串模糊匹配及其用例，并使用Python中Fuzzywuzzy库给出示例。

使用FuzzyWuzzy合并酒店房型

每个酒店都有自己的命名方法来命名它的房间，在线旅行社(OTA)也是如此。例如，同一家酒店的一间客房Expedia将之称为“Studio, 1 King Bed with Sofa Bed, Corner”，Booking.com（缤客）则简单地将其显示为“Corner King Studio”。不能说有谁错了，但是当我们想要比较OTA之间的房价时，或者一个OTA希望确保另一个OTA遵循费率平价协议时（rate parity agreement），这可能会导致混乱。换句话说，为了能够比较价格，我们必须确保我们进行比较的东西是同一类型的。对于价格比较网站和应用程序来说，最令人头条的问题之一就是试图弄清楚两个项目（比如酒店房间）是否是同一事物。

Fuzzywuzzy是一个Python库，使用编辑距离（Levenshtein Distance）来计算序列之间的差异。为了演示，我创建了自己的数据集，也就是说，对于同一酒店物业，我从Expedia拿一个房间类型，比如说“Suite, 1 King Bed (Parlor)”，然后我将它与Booking.com中的同类型房间匹配，即“King Parlor Suite”。只要有一点经验，大多数人都会知道他们是一样的。按照这种方法，我创建了一个包含100多对房间类型的小数据集，可以访问Github下载。

我们使用这个数据集测试Fuzzywuzzy的做法。换句话说，我们使用Fuzzywuzzy来匹配两个数据源之间的记录。

import pandas as pd df = pd. read_csv ( '../input/room_type.csv' ) df. head ( 10 ) import pandas as pd

df = pd.read_csv(’…/input/room_type.csv’)
df.head(10)

import pandas as pd

df = pd.read_csv(’…/input/room_type.csv’)
df.head(10)

有几种方法可以比较Fuzzywuzzy中的两个字符串，让我们一个一个地进行尝试。

ratio ，按顺序比较整个字符串的相似度

from fuzzywuzzy import fuzz fuzz. ratio ( 'Deluxe Room, 1 King Bed' , 'Deluxe King Room' ) from fuzzywuzzy import fuzz

fuzz.ratio(‘Deluxe Room, 1 King Bed’, ‘Deluxe King Room’)

from fuzzywuzzy import fuzz

fuzz.ratio(‘Deluxe Room, 1 King Bed’, ‘Deluxe King Room’)

返回结果时62，它告诉我们“Deluxe Room, 1 King Bed”和“Deluxe King Room”的相似度约62%。

fuzz. ratio ( 'Traditional Double Room, 2 Double Beds' , 'Double Room with Two Double Beds' ) fuzz. ratio ( 'Room, 2 Double Beds (19th to 25th Floors)' , 'Two Double Beds - Location Room (19th to 25th Floors)' ) fuzz.ratio('Traditional Double Room, 2 Double Beds','Double Room with Two Double Beds') fuzz.ratio('Room, 2 Double Beds (19th to 25th Floors)','Two Double Beds - Location Room (19th to 25th Floors)')

fuzz.ratio('Traditional Double Room, 2 Double Beds','Double Room with Two Double Beds')
fuzz.ratio('Room, 2 Double Beds (19th to 25th Floors)','Two Double Beds - Location Room (19th to 25th Floors)')

“Traditional Double Room, 2 Double Beds”和“Double Room with Two Double Beds”的相似度约69%。“Room, 2 Double Beds (19th to 25th Floors)”和“Two Double Beds — Location Room (19th to 25th Floors)”相似度约74%。显然效果不怎么样。事实证明，简单的方法对于词序，缺失或多余词语以及其他类似问题的微小差异太过敏感。

partial_ratio，比较部分字符串的相似度

我们仍在使用相同的数据对：

fuzz. partial_ratio ( 'Deluxe Room, 1 King Bed' , 'Deluxe King Room' ) fuzz. partial_ratio ( 'Traditional Double Room, 2 Double Beds' , 'Double Room with Two Double Beds' ) fuzz. partial_ratio ( 'Room, 2 Double Beds (19th to 25th Floors)' , 'Two Double Beds - Location Room (19th to 25th Floors)' ) fuzz.partial_ratio('Deluxe Room, 1 King Bed','Deluxe King Room') fuzz.partial_ratio('Traditional Double Room, 2 Double Beds','Double Room with Two Double Beds') fuzz.partial_ratio('Room, 2 Double Beds (19th to 25th Floors)','Two Double Beds - Location Room (19th to 25th Floors)')

fuzz.partial_ratio('Deluxe Room, 1 King Bed','Deluxe King Room')
fuzz.partial_ratio('Traditional Double Room, 2 Double Beds','Double Room with Two Double Beds')
fuzz.partial_ratio('Room, 2 Double Beds (19th to 25th Floors)','Two Double Beds - Location Room (19th to 25th Floors)')

返回依次69、83、63。对于我的数据集来说，比较部分字符串并不能带来更好的整体效果。让我们尝试下一个。

token_sort_ratio，忽略单词顺序

fuzz. token_sort_ratio ( 'Deluxe Room, 1 King Bed' , 'Deluxe King Room' ) fuzz. token_sort_ratio ( 'Traditional Double Room, 2 Double Beds' , 'Double Room with Two Double Beds' ) fuzz. token_sort_ratio ( 'Room, 2 Double Beds (19th to 25th Floors)' , 'Two Double Beds - Location Room (19th to 25th Floors)' ) fuzz.token_sort_ratio('Deluxe Room, 1 King Bed','Deluxe King Room') fuzz.token_sort_ratio('Traditional Double Room, 2 Double Beds','Double Room with Two Double Beds') fuzz.token_sort_ratio('Room, 2 Double Beds (19th to 25th Floors)','Two Double Beds - Location Room (19th to 25th Floors)')

fuzz.token_sort_ratio('Deluxe Room, 1 King Bed','Deluxe King Room')
fuzz.token_sort_ratio('Traditional Double Room, 2 Double Beds','Double Room with Two Double Beds')
fuzz.token_sort_ratio('Room, 2 Double Beds (19th to 25th Floors)','Two Double Beds - Location Room (19th to 25th Floors)')

返回依次84、78、83。这是迄今为止最好的。

token_set_ratio，去重子集匹配

它与token_sort_ratio类似，但更加灵活。

fuzz. token_set_ratio ( 'Deluxe Room, 1 King Bed' , 'Deluxe King Room' ) fuzz. token_set_ratio ( 'Traditional Double Room, 2 Double Beds' , 'Double Room with Two Double Beds' ) fuzz. token_set_ratio ( 'Room, 2 Double Beds (19th to 25th Floors)' , 'Two Double Beds - Location Room (19th to 25th Floors)' ) fuzz.token_set_ratio('Deluxe Room, 1 King Bed','Deluxe King Room') fuzz.token_set_ratio('Traditional Double Room, 2 Double Beds','Double Room with Two Double Beds') fuzz.token_set_ratio('Room, 2 Double Beds (19th to 25th Floors)','Two Double Beds - Location Room (19th to 25th Floors)')

fuzz.token_set_ratio('Deluxe Room, 1 King Bed','Deluxe King Room')
fuzz.token_set_ratio('Traditional Double Room, 2 Double Beds','Double Room with Two Double Beds')
fuzz.token_set_ratio('Room, 2 Double Beds (19th to 25th Floors)','Two Double Beds - Location Room (19th to 25th Floors)')

返回依次100、78、97。看来token_set_ratio最适合我的数据。根据这一发现，将token_set_ratio应用到整个数据集。

def get_ratio ( row ) : name1 = row [ 'Expedia' ] name2 = row [ 'Booking.com' ] return fuzz. token_set_ratio ( name1, name2 ) rated = df. apply ( get_ratio, axis= 1 ) rated. head ( 10 ) greater_than_70_percent = df [ rated > 70 ] greater_than_70_percent. count () len ( greater_than_70_percent ) / len ( df ) def get_ratio(row): name1 = row['Expedia'] name2 = row['Booking.com'] return fuzz.token_set_ratio(name1, name2)

rated = df.apply(get_ratio, axis=1)
rated.head(10)

greater_than_70_percent = df[rated > 70]
greater_than_70_percent.count()
len(greater_than_70_percent) / len(df)

def get_ratio(row):

name1 = row[‘Expedia’]

name2 = row[‘Booking.com’]

return fuzz.token_set_ratio(name1, name2)

rated = df.apply(get_ratio, axis=1)
rated.head(10)

greater_than_70_percent = df[rated > 70]
greater_than_70_percent.count()
len(greater_than_70_percent) / len(df)

当设定相似度> 70时，超过90％的房间对超过这个匹配分数。还很不错！上面只是做了2个文本间的相似度比较，如果存在多个如何处理？可以使用库中提供的 Process类：

用来返回模糊匹配的字符串和相似度。

>>> choices = ["Atlanta Falcons", "New York Jets", "New York Giants", "Dallas Cowboys"] >>> process.extract("new york jets", choices, limit=2) [('New York Jets', 100), ('New York Giants', 78)] >>> process.extractOne("cowboys", choices) ("Dallas Cowboys", 90) >>> choices = ["Atlanta Falcons", "New York Jets", "New York Giants", "Dallas Cowboys"] >>> process.extract("new york jets", choices, limit=2) [('New York Jets', 100), ('New York Giants', 78)] >>> process.extractOne("cowboys", choices) ("Dallas Cowboys", 90)

>>> choices = ["Atlanta Falcons", "New York Jets", "New York Giants", "Dallas Cowboys"]
    >>> process.extract("new york jets", choices, limit=2)
        [('New York Jets', 100), ('New York Giants', 78)]
    >>> process.extractOne("cowboys", choices)
        ("Dallas Cowboys", 90)

FuzzyWuzzy在中文场景下的使用

FuzzyWuzzy支持对中文进行比较：

from fuzzywuzzy import fuzz from fuzzywuzzy import process print ( fuzz. ratio ( "数据挖掘" , "数据挖掘工程师" )) title_list = [ "数据分析师" , "数据挖掘工程师" , "大数据开发工程师" , "机器学习工程师" , "算法工程师" , "数据库管理" , "商业分析师" , "数据科学家" , "首席数据官" , "数据产品经理" , "数据运营" , "大数据架构师" ] print ( process. extractOne ( "数据挖掘" , title_list )) from fuzzywuzzy import fuzz from fuzzywuzzy import process

print(fuzz.ratio(“数据挖掘”, “数据挖掘工程师”))

title_list = [“数据分析师”, “数据挖掘工程师”, “大数据开发工程师”, “机器学习工程师”,
“算法工程师”, “数据库管理”, “商业分析师”, “数据科学家”, “首席数据官”,
“数据产品经理”, “数据运营”, “大数据架构师”]

print(process.extractOne(“数据挖掘”, title_list))

from fuzzywuzzy import fuzz

from fuzzywuzzy import process

print(fuzz.ratio(“数据挖掘”, “数据挖掘工程师”))

title_list = [“数据分析师”, “数据挖掘工程师”, “大数据开发工程师”, “机器学习工程师”,
“算法工程师”, “数据库管理”, “商业分析师”, “数据科学家”, “首席数据官”,
“数据产品经理”, “数据运营”, “大数据架构师”]

print(process.extractOne(“数据挖掘”, title_list))

仔细查看代码，还是存在的问题：

• FuzzWuzzy并不会针对中文进行分词
• 也没有对中文的一些停用词进行过滤

改进方案，处理前进行中文处理：

• 繁简转换
• 中文分词
• 去除停用词

参考链接：

• https://github.com/seatgeek/fuzzywuzzy
• https://www.kaggle.com/leandrodoze/fuzzy-string-matching-with-hotel-rooms

相关文章:

1. Google词向量该工具Word2Vec
2. Python地图可视化之Folium
3. Skyline实战：CentOS 7部署
4. ElasticSearch 实现经纬度附近搜索
5. 自然语言处理工具包推荐

python

标签：King,FuzzyWuzzy,ratio,Room,Python,Double,Beds,fuzz,字符串
来源： https://blog.csdn.net/stay_foolish12/article/details/113109000