Develop Reference

Custom fuzzy search

Good afternoon everyone. There is a parser, which the text receives as input, it is necessary to determine the category in this text by keywords before writing to the database. At the moment the algorithm is as follows:
I bring everything to lowercase
I remove the excess with a regular expression ([^a-za-z])
I delete all words that consist of less than 4 characters (category name 4+ characters)
I break the line into words
I remove duplicate words
In the cycle, I go through all the categories and use the Levenshtein (fuzzywuzzy library) algorithm to determine the most suitable one.
There will be about 100 categories, and each category will have 100 keywords.
What tools or algorithms can be used to more effectively solve the problem?
def detect_category(post):
tech_categories = ["macbook", "ноутбук", "процессор", "экран", "оперативная память", "видеокарта", "аккумулятор"]
result = re.sub(r'[^a-zа-я ]+', '', post)
post_without_small_words = ' '.join([w for w in result.split() if len(w) > 3])
words = post_without_small_words.split(' ')
unique_words = sorted(set(words), key=words.index)
good = []
for tech_category in tech_categories:
ratios = process.extract(tech_category, unique_words)
for ration in ratios:
if ration[1] > 90:
# print(ration)
good.append(ration)
return good

Is there a way to do fuzzy string matching for words on string?

I want to do fuzzy matching on string with words.
The target string could be like.
"Hello, I am going to watch a film today."
where the words I want to search are.
"flim toda".
This hopefully should return "film today" as a search result.
I have used this method but it seems to be working only with one word.
import difflib
def matches(large_string, query_string, threshold):
words = large_string.split()
matched_words = []
for word in words:
s = difflib.SequenceMatcher(None, word, query_string)
match = ''.join(word[i:i+n] for i, j, n in s.get_matching_blocks() if n)
if len(match) / float(len(query_string)) >= threshold:
matched_words.append(match)
return matched_words
large_string = "Hello, I am going to watch a film today"
query_string = "film"
print(list(matches(large_string, query_string, 0.8)))
This only works with one word and it returns when there is little noise.
Is there any way to do such fuzzy matching with words?

The feature you are thinking of is called "query suggestion" and does rely on spell checking, but it relies on markov chains built out of search engine query log.
That being said, you use an approach similar to the one described in this answer: https://stackoverflow.com/a/58166648/140837

You can simply use Fuzzysearch, please see the example below;
from fuzzysearch import find_near_matches
text_string = "Hello, I am going to watch a film today."
matches = find_near_matches('flim toda', text_string, max_l_dist=2)
print([my_string[m.start:m.end] for m in matches])
This will give you the desired output.
['film toda']
Please note that you can give a value for max_l_dist parameter based on how much you are going to tolerate.

auto-correct the words from the list in python

I want to auto-correct the words which are in my list.
Say I have a list
kw = ['tiger','lion','elephant','black cat','dog']
I want to check if these words appeared in my sentence. If they are wrongly spelled I want to correct them. I don't intend to touch other words except from the given list.
Now I have list of str
s = ["I saw a tyger","There are 2 lyons","I mispelled Kat","bulldogs"]
Expected output:
['tiger','lion',None,'dog']
My Efforts:
import difflib
op = [difflib.get_close_matches(i,kw,cutoff=0.5) for i in s]
print(op)
My Output:
[[], [], [], ['dog']]
The problem with above code is I want to compare entire sentence and my kw list can have more than 1 word(upto 4-5 words).
If I lower the cutoff value it starts returning the words which is should not.
So even if I plan to create bigrams, trigrams from given sentence it would consume a lot of time.
So is there way to implement this?
I have explored few more libraries like autocorrect, hunspell etc. but no success.

You could implement something based of levenshtein distance.
It's interesting to note elasticsearch's implementation: https://www.elastic.co/guide/en/elasticsearch/guide/master/fuzziness.html
Clearly, bieber is a long way from beaver—they are too far apart to be
considered a simple misspelling. Damerau observed that 80% of human
misspellings have an edit distance of 1. In other words, 80% of
misspellings could be corrected with a single edit to the original
string.
Elasticsearch supports a maximum edit distance, specified with the
fuzziness parameter, of 2.
Of course, the impact that a single edit has on a string depends on
the length of the string. Two edits to the word hat can produce mad,
so allowing two edits on a string of length 3 is overkill. The
fuzziness parameter can be set to AUTO, which results in the following
maximum edit distances:
0 for strings of one or two characters
1 for strings of three, four, or five characters
2 for strings of more than five characters
I like to use pyxDamerauLevenshtein myself.
pip install pyxDamerauLevenshtein
So you could do a simple implementation like:
keywords = ['tiger','lion','elephant','black cat','dog']
from pyxdameraulevenshtein import damerau_levenshtein_distance
def correct_sentence(sentence):
new_sentence = []
for word in sentence.split():
budget = 2
n = len(word)
if n < 3:
budget = 0
elif 3 <= n < 6:
budget = 1
if budget:
for keyword in keywords:
if damerau_levenshtein_distance(word, keyword) <= budget:
new_sentence.append(keyword)
break
else:
new_sentence.append(word)
else:
new_sentence.append(word)
return " ".join(new_sentence)
Just make sure you use a better tokenizer or this will get messy, but you get the point. Also note that this is unoptimized, and will be really slow with a lot of keywords. You should implement some kind of bucketing to not match all words with all keywords.

Here is one way using difflib.SequenceMatcher. The SequenceMatcher class allows you to measure sentence similarity with its ratio method, you only need to provide a suitable threshold in order to keep words with a ratio that falls above the given threshold:
def find_similar_word(s, kw, thr=0.5):
from difflib import SequenceMatcher
out = []
for i in s:
f = False
for j in i.split():
for k in kw:
if SequenceMatcher(a=j, b=k).ratio() > thr:
out.append(k)
f = True
if f:
break
if f:
break
else:
out.append(None)
return out
Output
find_similar_word(s, kw)
['tiger', 'lion', None, 'dog']

Although this is slightly different from your expected output (it is a list of list instead of a list of string) I thing it is a step in the right direction. The reason I chose this method, is so that you can have multiple corrections per sentence. That is why I added another example sentence.
import difflib
import itertools
kw = ['tiger','lion','elephant','black cat','dog']
s = ["I saw a tyger","There are 2 lyons","I mispelled Kat","bulldogs", "A tyger is different from a doog"]
op = [[difflib.get_close_matches(j,kw,cutoff=0.5) for j in i.split()] for i in s]
op = [list(itertools.chain(*o)) for o in op]
print(op)
The output is generate is:
[['tiger'], ['lion'], [], ['dog'], ['tiger', 'dog']]
The trick is to split all the sentences along the whitespaces.

Find best substring match

I'm looking for a library or a method using existing libraries( difflib, fuzzywuzzy, python-levenshtein) to find the closest match of a string (query) in a text (corpus)
I've developped a method based on difflib, where I split my corpus into ngrams of size n (length of query).
import difflib
from nltk.util import ngrams
def get_best_match(query, corpus):
ngs = ngrams( list(corpus), len(query) )
ngrams_text = [''.join(x) for x in ngs]
return difflib.get_close_matches(query, ngrams_text, n=1, cutoff=0)
it works as I want when the difference between the query and the matched string are just character replacements.
query = "ipsum dolor"
corpus = "lorem 1psum d0l0r sit amet"
match = get_best_match(query, corpus)
# match = "1psum d0l0r"
But when the difference is character deletion, it is not.
query = "ipsum dolor"
corpus = "lorem 1psum dlr sit amet"
match = get_best_match(query, corpus)
# match = "psum dlr si"
# expected_match = "1psum dlr"
Is there a way to get a more flexible result size ( as for expected_match ) ?
EDIT 1:
The actual use of this script is to match queries (strings) with a
messy ocr output.
As I said in the question, the ocr can confound characters, and even miss them.
If possible consider also the case when a space is missing between words.
A best match, is the one that does not include characters from other words than those on the query.
EDIT 2:
The solution I use now is to extend the ngrams with (n-k)-grams for k = {1,2,3} to prevent 3 deletions. It's much better than the first version, but not efficient in terms of speed, as we have more than 3 times the number of ngrams to check. It is also a non generalizable solution.

This function finds best matching substring of variable length.
The implementation considers the corpus as one long string, hence avoiding your concerns with spaces and unseparated words.
Code summary:
1. Scan the corpus for match values in steps of size step to find the approximate location of highest match value, pos.
2. Find the substring in the vicinity of pos with the highest match value, by adjusting the left/right positions of the substring.
from difflib import SequenceMatcher
def get_best_match(query, corpus, step=4, flex=3, case_sensitive=False, verbose=False):
"""Return best matching substring of corpus.
Parameters
----------
query : str
corpus : str
step : int
Step size of first match-value scan through corpus. Can be thought of
as a sort of "scan resolution". Should not exceed length of query.
flex : int
Max. left/right substring position adjustment value. Should not
exceed length of query / 2.
Outputs
-------
output0 : str
Best matching substring.
output1 : float
Match ratio of best matching substring. 1 is perfect match.
"""
def _match(a, b):
"""Compact alias for SequenceMatcher."""
return SequenceMatcher(None, a, b).ratio()
def scan_corpus(step):
"""Return list of match values from corpus-wide scan."""
match_values = []
m = 0
while m + qlen - step <= len(corpus):
match_values.append(_match(query, corpus[m : m-1+qlen]))
if verbose:
print(query, "-", corpus[m: m + qlen], _match(query, corpus[m: m + qlen]))
m += step
return match_values
def index_max(v):
"""Return index of max value."""
return max(range(len(v)), key=v.__getitem__)
def adjust_left_right_positions():
"""Return left/right positions for best string match."""
# bp_* is synonym for 'Best Position Left/Right' and are adjusted
# to optimize bmv_*
p_l, bp_l = [pos] * 2
p_r, bp_r = [pos + qlen] * 2
# bmv_* are declared here in case they are untouched in optimization
bmv_l = match_values[p_l // step]
bmv_r = match_values[p_l // step]
for f in range(flex):
ll = _match(query, corpus[p_l - f: p_r])
if ll > bmv_l:
bmv_l = ll
bp_l = p_l - f
lr = _match(query, corpus[p_l + f: p_r])
if lr > bmv_l:
bmv_l = lr
bp_l = p_l + f
rl = _match(query, corpus[p_l: p_r - f])
if rl > bmv_r:
bmv_r = rl
bp_r = p_r - f
rr = _match(query, corpus[p_l: p_r + f])
if rr > bmv_r:
bmv_r = rr
bp_r = p_r + f
if verbose:
print("\n" + str(f))
print("ll: -- value: %f -- snippet: %s" % (ll, corpus[p_l - f: p_r]))
print("lr: -- value: %f -- snippet: %s" % (lr, corpus[p_l + f: p_r]))
print("rl: -- value: %f -- snippet: %s" % (rl, corpus[p_l: p_r - f]))
print("rr: -- value: %f -- snippet: %s" % (rl, corpus[p_l: p_r + f]))
return bp_l, bp_r, _match(query, corpus[bp_l : bp_r])
if not case_sensitive:
query = query.lower()
corpus = corpus.lower()
qlen = len(query)
if flex >= qlen/2:
print("Warning: flex exceeds length of query / 2. Setting to default.")
flex = 3
match_values = scan_corpus(step)
pos = index_max(match_values) * step
pos_left, pos_right, match_value = adjust_left_right_positions()
return corpus[pos_left: pos_right].strip(), match_value
Example:
query = "ipsum dolor"
corpus = "lorem i psum d0l0r sit amet"
match = get_best_match(query, corpus, step=2, flex=4)
print(match)
('i psum d0l0r', 0.782608695652174)
Some good heuristic advice is to always keep step < len(query) * 3/4, and flex < len(query) / 3. I also added case sensitivity, in case that's important. It works quite well when you start playing with the step and flex values. Small step values gives better results but takes longer to compute. flex governs how flexible the length of the resulting substring is allowed to be.
Important to note: This will only find the first best match, so if there are multiple equally good matches, only the first will be returned. To allow for multiple matches, change index_max() to return a list of indices for the n highest values of the input list, and loop over adjust_left_right_positions() for values in that list.

The main path to a solution uses finite state automata (FSA) of some kind. If you want a detailed summary of the topic, check this dissertation out (PDF link). Error-based models (including Levenshtein automata and transducers, the former of which Sergei mentioned) are valid approaches to this. However, stochastic models, including various types of machine learning approaches integrated with FSAs, are very popular at the moment.
Since we are looking at edit distances (effectively misspelled words), the Levenshtein approach is good and relatively simple. This paper (as well as the dissertation; also PDF) give a decent outline of the basic idea and it also explicitly mentions the application to OCR tasks. However, I will review some of the key points below.
The basic idea is that you want to build an FSA that computes both the valid string as well as all strings up to some error distance (k). In the general case, this k could be infinite or the size of the text, but this is mostly irrelevant for OCR (if your OCR could even potentially return bl*h where * is the rest of the entire text, I would advise finding a better OCR system). Hence, we can restrict regex's like bl*h from the set of valid answers for the search string blah. A general, simple and intuitive k for your context is probably the length of the string (w) minus 2. This allows b--h to be a valid string for blah. It also allows bla--h, but that's okay. Also, keep in mind that the errors can be any character you specify, including spaces (hence 'multiword' input is solvable).
The next basic task is to set up a simple weighted transducer. Any of the OpenFST Python ports can do this (here's one). The logic is simple: insertions and deletions increment the weight while equality increments the index in the input string. You could also just hand code it as the guy in Sergei's comment link did.
Once you have the weights and associated indexes of the weights, you just sort and return. The computational complexity should be O(n(w+k)), since we will look ahead w+k characters in the worst case for each character (n) in the text.
From here, you can do all sorts of things. You could convert the transducer to a DFA. You could parallelize the system by breaking the text into w+k-grams, which are sent to different processes. You could develop a language model or confusion matrix that defines what common mistakes exist for each letter in the input set (and thereby restrict the space of valid transitions and the complexity of the associated FSA). The literature is vast and still growing so there are probably as many modifications as there are solutions (if not more).
Hopefully that answers some of your questions without giving any code.

I would try to build a regular expression template from the query string. The template could then be used to search the corpus for substrings that are likely to match the query. Then use difflib or fuzzywuzzy to check if the substring does match the query.
For example, a possible template would be to match at least one of the first two letters of the query, at least one of the last two letters of the query, and have approximately the right number of letters in between:
import re
query = "ipsum dolor"
corpus = ["lorem 1psum d0l0r sit amet",
"lorem 1psum dlr sit amet",
"lorem ixxxxxxxr sit amet"]
first_letter, second_letter = query[:2]
minimum_gap, maximum_gap = len(query) - 6, len(query) - 3
penultimate_letter, ultimate_letter = query[-2:]
fmt = '(?:{}.|.{}).{{{},{}}}(?:{}.|.{})'.format
pattern = fmt(first_letter, second_letter,
minimum_gap, maximum_gap,
penultimate_letter, ultimate_letter)
#print(pattern) # for debugging pattern
m = difflib.SequenceMatcher(None, "", query, False)
for c in corpus:
for match in re.finditer(pattern1, c, re.IGNORECASE):
substring = match.group()
m.set_seq1(substring)
ops = m.get_opcodes()
# EDIT fixed calculation of the number of edits
#num_edits = sum(1 for t,_,_,_,_ in ops if t != 'equal')
num_edits = sum(max(i2-i1, j2-j1) for op,i1,i2,j1,j2 in ops if op != 'equal' )
print(num_edits, substring)
Output:
3 1psum d0l0r
3 1psum dlr
9 ixxxxxxxr
Another idea is to use the characteristics of the ocr when building the regex. For example, if the ocr always gets certain letters correct, then when any of those letters are in the query, use a few of them in the regex. Or if the ocr mixes up '1', '!', 'l', and 'i', but never substitutes something else, then if one of those letters is in the query, use [1!il] in the regex.

What is a good strategy to group similar words?

Say I have a list of movie names with misspellings and small variations like this -
"Pirates of the Caribbean: The Curse of the Black Pearl"
"Pirates of the carribean"
"Pirates of the Caribbean: Dead Man's Chest"
"Pirates of the Caribbean trilogy"
"Pirates of the Caribbean"
"Pirates Of The Carribean"
How do I group or find such sets of words, preferably using python and/or redis?

Have a look at "fuzzy matching". Some great tools in the thread below that calculates similarities between strings.
I'm especially fond of the difflib module
>>> get_close_matches('appel', ['ape', 'apple', 'peach', 'puppy'])
['apple', 'ape']
>>> import keyword
>>> get_close_matches('wheel', keyword.kwlist)
['while']
>>> get_close_matches('apple', keyword.kwlist)
[]
>>> get_close_matches('accept', keyword.kwlist)
['except']
https://stackoverflow.com/questions/682367/good-python-modules-for-fuzzy-string-comparison

You might notice that similar strings have large common substring, for example:
"Bla bla bLa" and "Bla bla bRa" => common substring is "Bla bla ba" (notice the third word)
To find common substring you may use dynamic programming algorithm. One of algorithms variations is Levenshtein distance (distance between most similar strings is very small, and between more different strings distance is bigger) - http://en.wikipedia.org/wiki/Levenshtein_distance.
Also for quick performance you may try to adapt Soundex algorithm - http://en.wikipedia.org/wiki/Soundex.
So after calculating distance between all your strings, you have to clusterize them. The most simple way is k-means (but it needs you to define number of clusters). If you actually don't know number of clusters, you have to use hierarchical clustering. Note that number of clusters in your situation is number of different movies titles + 1(for totally bad spelled strings).

I believe there is in fact two distinct problems.
The first is spell correction. You can have one in Python here
http://norvig.com/spell-correct.html
The second is more functional. Here is what I'd do after the spell correction. I would make a relation function.
related( sentence1, sentence2 ) if and only if sentence1 and sentence2 have rare common words. By rare, I mean words different than (The, what, is, etc...). You can take a look at the TF/IDF system to determine if two document are related using their words. Just googling a bit I found this:
https://code.google.com/p/tfidf/

To add another tip to Fredrik's answer, you could also get inspired from search engines like code, such as this one :
def dosearch(terms, searchtype, case, adddir, files = []):
found = []
if files != None:
titlesrch = re.compile('>title<.*>/title<')
for file in files:
title = ""
if not (file.lower().endswith("html") or file.lower().endswith("htm")):
continue
filecontents = open(BASE_DIR + adddir + file, 'r').read()
titletmp = titlesrch.search(filecontents)
if titletmp != None:
title = filecontents.strip()[titletmp.start() + 7:titletmp.end() - 8]
filecontents = remove_tags(filecontents)
filecontents = filecontents.lstrip()
filecontents = filecontents.rstrip()
if dofind(filecontents, case, searchtype, terms) > 0:
found.append(title)
found.append(file)
return found
Source and more information: http://www.zackgrossbart.com/hackito/search-engine-python/
Regards,
Max

One approach would be to pre-process all the strings before you compare them: convert all to lowercase, standardize whitespace (eg, replace any whitespace with single spaces). If punctuation is not important to your end goal, you can remove all punctuation characters as well.
Levenshtein distance is commonly-used to determine similarity of a string, this should help you group strings which differ by small spelling errors.