字符串匹配 - 使用的最佳距离算法

Question

I have two dataframes, df1 and df2 , that have information about polling stations. The dataframes are of different lengths. Both dataframes have a column called ps_name , which is the name of the polling stations, and a column called district that indicates which district the polling stations are located.

I am trying to match strings on the ps_name column while blocking on the district column, so I can copy a geolocations (latitude and longitude) column on matches from df1 to df2 .

So far I've tried using jaro-winkler at threshold 0.88 to compare strings.

# Matched:
**df1:** AGRICULTURAL OFFICE ATTOCK (MALE) I (P)
**df2:** AGRICULTURAL OFFICE ATTOCK (MALE) (P)

# Did not match:
**df1:** govt girls high school peoples colony attock ii
**df2:** high school peoples colony attock ii

What string distance algorithm should I be using ? I've tried jaro-winkler and was also considering smith-waterman .

Answer 1

One option is to use Levenshtein distance which is implemented in the package fuzzywuzzy (or here ), the algorithm runs in O(n + d^2), where n is the length of the longer string and d is the edit distance.

Example:

from fuzzywuzzy import fuzz
fuzz.ratio('govt girls high school peoples colony attock ii','high school peoples colony attock ii') 
#87
fuzz.ratio('AGRICULTURAL OFFICE ATTOCK (MALE) I (P)', 'AGRICULTURAL OFFICE ATTOCK (MALE) (P)')
#97