Python：比较来自 csv 的行并将 pdf 布局比较的相同结果组合在一起

Question

I am trying to find a way to compare the layout of different pdf files. Using tesseract, I am able to export to a CSV file the following data of specific keywords.

Consider this generated csv file with the following content, displaying the left and top coordinates of each keyword, as well as the keyword and the file name:

Left,Top,Text,File
118,174,INVOICE,file0
117,333,INVOICE,file0
119,525,BILLED,file0
119,1554,INVOICE,file0
322,1880,invoice,file0
118,174,INVOICE,file1
117,333,INVOICE,file1
119,525,BILLED,file1
119,1554,INVOICE,file1
322,1880,invoice,file1
1112,185,Invoice,file2
113,219,Invoice,file2
1112,212,Invoice,file3
113,219,Invoice,file3
113,217,Invoice,file3
118,174,INVOICE,file4
117,333,INVOICE,file4
119,525,BILLED,file4
119,1554,INVOICE,file4
322,1884,invoice,file4

My initial idea is to concatenate the first 3 columns and compare them to the other rows. I am able to obtain which files match for each keyword. But I am unable to obtain which files have an overall matching layout of more than 80% for example.

Here is my code so far:

import pandas as pd
import itertools

Loop over csv to get position and text of keywords

with open('data.csv') as file:
    results = []
    file_names = []
    for row in file:
        columns = row.split(',')
        data = columns[0] + columns[1] + columns[2]
        file_name = columns[3].rstrip()
        results = results + [data]
        file_names.append(file_name)

Get and print indices of the matches

indices = [] 
for a, b in itertools.combinations(results, 2):
    if a == b:
        indices = indices + [[i -1 for i, x in enumerate(results) if x == a]]
print("Indices: ", indices)

Print:

Indices:  [[0, 5, 15], [0, 5, 15], [1, 6, 16], [1, 6, 16], [2, 7, 17], [2, 7, 17], [3, 8, 18], [3, 8, 18], [4, 9], [0, 5, 15], [1, 6, 16], [2, 7, 17], [3, 8, 18], [11, 13]]

Get and print filenames with match

dataset = pd.read_csv('data.csv', sep=',')
identical_files = []
for indice in indices:
    file_matches = []
    for i in indice:
        file_matches.append(dataset.iloc[i, -1])
    identical_files.append(file_matches)
print("Identical files: ", identical_files)

Print:

Identical files:  [['file0', 'file1', 'file4'], ['file0', 'file1', 'file4'], ['file0', 'file1', 'file4'], ['file0', 'file1', 'file4'], ['file0', 'file1', 'file4'], ['file0', 'file1', 'file4'], ['file0', 'file1', 'file4'], ['file0', 'file1', 'file4'], ['file0', 'file1'], ['file0', 'file1', 'file4'], ['file0', 'file1', 'file4'], ['file0', 'file1', 'file4'], ['file0', 'file1', 'file4'], ['file2', 'file3']]

So I am able to print the identical files, however, after many attempts I am struggling to figure out the logic to identify which files have an identical layout and should, therefore, be grouped together.

Based on this data, the output should be something like this:

[
  [file0, file1, file4],
  [file2, file3]
]

I am still new to Python so I hope I have made myself clear.

Answer 1

I am not sure if this is exactly what you need, but try this:

res = (df.groupby(by=['Left', 'Top', 'Text'])
       .agg(files = pd.NamedAgg(column="File", aggfunc=', '.join)))
print(res)

This will give me

                                 files
Left Top  Text                        
1112 185  Invoice                file2
     212  Invoice                file3
113  217  Invoice                file3
     219  Invoice         file2, file3
117  333  INVOICE  file0, file1, file4
118  174  INVOICE  file0, file1, file4
119  1554 INVOICE  file0, file1, file4
     525  BILLED   file0, file1, file4
322  1880 invoice         file0, file1
     1884 invoice                file4

which are for all combinatiosn of "Left", "Top", and "Text" the files which have the same combinations.

Does that help?