Comparing two strings with low/no consistency

Question

I have two strings

a = 'Test - 4567: Controlling_robotic_hand_with_Arduino_uno'
b = 'Controlling robotic hand'

I need to check if they match and print out the result accordingly. As b is the string I want checked in a, the result should print out 'Match' or 'Mis-match' accordingly. The code shoule not depend on the '_' in a, as they can be '-' or spaces as well. I have tried using fuzzywuzzy library and the fuzzy.token_set_ratio to calculate the ratio. From observation, I chose a value of 95 to be convincing. I want to know if there is another way to check this without using fuzzywuzzy, probably difflib. I tried using difflib and SequenceManager, but all I get is a word wise comparison and am unable to combine the result exactly.

I have tried the following code.

from fuzzywuzzy import fuzzy
a = 'Test - 4567: robotic_hand_with_Arduino_uno_controlling_pos0_pos1'
b = 'Controlling from pos0 to pos1'
ratio = fuzzy.token_set_ratio(a.lower(), b.lower())
if ratio >= 95:
    print('Match')
else:
    print('Mis-Match')

output

'Mis-Match'

This gives a score of 64 while all of controlling, pos0 and pos1 are in a and in b and should give a match instead.

I tried this as this doesn't depend on the '_' or '-' or spaces.

Answer 1

You can use gensim library implement MatchSemantic and write code like this as a function:

Initialization

---

if run the code for the first time a process-bar will go from 0% to 100% for downloading glove-wiki-gigaword-50 of the gensim and after that everything will be set and you can simply run the code

Code

---

from re import sub
import numpy as np
from gensim.utils import simple_preprocess
import gensim.downloader as api
from gensim.corpora import Dictionary
from gensim.models import TfidfModel
from gensim.similarities import SparseTermSimilarityMatrix, WordEmbeddingSimilarityIndex, SoftCosineSimilarity

def MatchSemantic(query_string, documents):
    stopwords = ['the', 'and', 'are', 'a']

    if len(documents) == 1: documents.append('')

    def preprocess(doc):
        # Tokenize, clean up input document string
        doc = sub(r'<img[^<>]+(>|$)', " image_token ", doc)
        doc = sub(r'<[^<>]+(>|$)', " ", doc)
        doc = sub(r'\[img_assist[^]]*?\]', " ", doc)
        doc = sub(r'http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\(\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+', " url_token ", doc)
        return [token for token in simple_preprocess(doc, min_len=0, max_len=float("inf")) if token not in stopwords]

    # Preprocess the documents, including the query string
    corpus = [preprocess(document) for document in documents]
    query = preprocess(query_string)

    # Load the model: this is a big file, can take a while to download and open
    glove = api.load("glove-wiki-gigaword-50")
    similarity_index = WordEmbeddingSimilarityIndex(glove)

    # Build the term dictionary, TF-idf model
    dictionary = Dictionary(corpus + [query])
    tfidf = TfidfModel(dictionary=dictionary)

    # Create the term similarity matrix.
    similarity_matrix = SparseTermSimilarityMatrix(similarity_index, dictionary, tfidf)

    query_tf = tfidf[dictionary.doc2bow(query)]

    index = SoftCosineSimilarity(
        tfidf[[dictionary.doc2bow(document) for document in corpus]],
        similarity_matrix)

    doc_similarity_scores = index[query_tf]

    # Output the sorted similarity scores and documents
    sorted_indexes = np.argsort(doc_similarity_scores)[::-1]

    for idx in sorted_indexes:
        if documents[idx] != '':
            if doc_similarity_scores[idx] > 0.0: print('Match')
            else: print('Mis-Match')

Usage

---

for example, we want to see if Fruit and Vegetables matches any of the sentences or items inside documents

Test:

query_string = 'Fruit and Vegetables'
documents = ['I have an apple in my basket', 'I have a car in my house']
MatchSemantic(query_string, documents)

---

so we know that the first item I have an apple in my basket has a semantical relation with Fruit and Vegetables so it prints Match and for the second item no relation will be found so it prints Mis-Match

output:

Match
Mis-Match

Answer 2

It looks like you are trying to use tools like fuzzywuzzy that were not really designed for that.

One possible approach to this problem could be to find how many tokens from the second text are present in the first text. This can be normalized by the total number of tokens in the second text. Then you can threshold to whatever value you deem fit.

One possible way of implementing this is the following:

1. Tokenize (i.e. convert to a list of tokens) the input texts a and b.
2. Collect each token list into a corresponding counter (i.e. some data structure for counting the which tokens are present).
3. Compute the intersection a_i_b of the tokens for a and b.
4. Compute some metric based on the the total occurrences of a_i_b (weight_a_i_b) and the total occurrences of b (weight_b). This final metric is a proxy of the "amount" of b contained into a. This could be a ratio or a difference and should use the fact that weight_a_i_b <= weight_b by construction.

The difference weight_b - weight_a_i_b results in a number between 0 and the number of tokens in b, which is also a direct measure of how many tokens from b are not found in a, hence 0 indicates perfect matching.

The ratio weight_a_i_b / weight_b results in a number between 0 and 1, with 1 meaning perfect matching and 0 meaning no matching.

The difference metric is probably more suited for small number of tokens and easier to interpret and threshold in a meaningful way (e.g. accepting a value below 2 means that at most one token from b is not present in a).

On the other hand the ratio is more standard and it is probably more suited for larger tokens lists sizes.

All this would translate into this code, leveraging collections.Counter() for the dealing with counting the tokens:

import collections


def contains_tokens(
        text_a,
        text_b,
        tokenize_kws=None,
        metric=lambda a, b, a_i_b: b - a_i_b):
    """Compute the ratio of `b` contained in `a`."""
    tokenize_kws = dict(tokenize_kws) if tokenize_kws is not None else {}
    counter_a = collections.Counter(tokenize(text_a, **tokenize_kws))
    counter_b = collections.Counter(tokenize(text_b, **tokenize_kws))
    counter_a_i_b = counter_a & counter_b
    weight_a = counter_total(counter_a)
    weight_b = counter_total(counter_b)
    weight_a_i_b = counter_total(counter_a_i_b)
    return metric(weight_a, weight_b, weight_a_i_b)

The first step, i.e. tokenization, is achieved with the following function. This is a bit primitive, but gets the job done for the your input. What it does essentially is to replace a number of special characters (ignores) into blanks, and then splits the string along the blanks, optionally excluding the tokens in a blacklist (excludes).

def tokenize(
        text,
        case_sensitive=False,
        ignores=('_', '-', ':', ',', '.', '?', '!'),
        excludes=('the', 'from', 'to')):
    """Tokenize a text, ignoring some characters and excluding some tokens."""
    if not case_sensitive:
        text = text.lower()
    for ignore in ignores:
        text = text.replace(ignore, ' ')
    for token in text.split():
        if token not in excludes:
            yield token

To count the total number of values in a counter the following function is used. However, for Python 3.10 and later, there is a build-in method Counter.total() which does the exact same.

def counter_total(counter):
    """Count the total number of values."""
    return sum(counter.values())

---

For the given input this becomes:

a = 'Test - 4567: Controlling_robotic_hand_with_Arduino_uno'
b = 'Controlling robotic hand'


# all tokens from `b` are in `a`
print(contains_tokens(a, b))
# 0

and

a = 'Test - 4567: robotic_hand_with_Arduino_uno_controlling_pos0_pos1'
b = 'Controlling from pos0 to pos1'


# all tokens from `b` are in `a`
print(contains_tokens(a, b))
# 0

a = 'Test - 4567: robotic_hand_with_Arduino_uno_controlling_pos0_pos1'
b = 'Controlling from pos0 to pos2'


# one token from `b` (`pos2`) not in `a`
print(contains_tokens(a, b))
# 1

---

Note that distance-based functions (like fuzz.token_set_ratio() or fuzz.partial_ratio()) cannot be used in this context because they will be sensitive to how much "noise" is present in the first text, e.g. if b = 'a b c', those tokens are contained equally in a = 'a c' as well as a = 'a b c d e f g h i', and any distance cannot account for that, most notably because distance functions are symmetric (i.e. f(a, b) = f(b, a)) while the function you are looking for is not (i.e. f(a, b) != f(b, a)).