TextNormalizer

TextNormalizer is a string normalizer that uses SentenceTransformers as a backbone to obtain vector representations of sentences.
It's designed for repeated normalization task against a large corpus of strings.

The main contribution of TextNormalizer is to gain time by eliminating the need to compute the normalized strings embeddings every time.

Setup

pip install t-normalizer

Usage

1. Create an instance of TextNormalizer, can be initialized with a SentenceTransformer model or a SentenceTransformer model path.
2. Obtain the vector representation of the normalized string with .fit method.
3. Transform the string to the most similar normalized form using the .transform method.

from textnormalizer import TextNormalizer

normalizer = TextNormalizer()

normalized_text = ['senior software engineer', 'solutions architect', 'junior software developer']
to_normalize = ['experienced software engineer', 'software architect', 'entry level software engineer']

normalizer.fit(normalized_text)
transformed = normalizer.transform(to_normalize)

Serialization

The model along with the normalized strings and their vector representations can be saved and loaded with .save and .load methods.

# save
normalizer.save('path/to/model')

# load
model = TextNormalizer.load('path/to/model')