tfidf = TfidfVectorizer().fit_transform(documents)
min_percent = int(min_percent) / 100.0
row_length = len(documents)

result = {}

for i in range(row_length):
    row = linear_kernel(tfidf[i:i+1], tfidf).flatten()
    for n in range(row_length):
        if i != n and row[n] >= min_percent:
            if not result.get(i):
                result[i] = []
            result[i].append((n, row[n]))

# Postprocessing
clean_result = deepcopy(result)
for key, values in result.iteritems():
    if key in clean_result:
        for value in values:
            if value in clean_result[key] and value[0] in clean_result:
                for value2 in clean_result[value[0]]:
                    if value2[0] == key:
                        clean_result[value[0]].remove(value2)
                if not clean_result[value[0]]:
                    clean_result.pop(value[0])