A Generic Sentence Embedding Library

In natural language processing, we need to encode text data. In the past, we mostly use encoders such as one-hot, term frequency, or TF-IDF (normalized term frequency). There are many challenges with these techniques. In the recent years, the latest advancements give us opportunity to encode sentences or words in more meaningful format. The word2vec model was one of them. Plus, the BERT language model provides us a powerful sentence encoders.

The sentence encoding is an important step of many NLP projects. Plus, we believe that a flexible sent2vec library is needed to build a prototype fast. That is why we have initiated this project. In the early releases, you will have access to the standard encoders. We will add more curated techniques in the later releases. Hope you can use this library in your exciting NLP projects. Note that we can use encoding, embedding, or vectorizing interchangeably in this context.

Library

The package requires the following libraries:

• gensim
• numpy
• pandas
• spacy
• transformers
• torch

This package is developed to make the prototyping become faster. That is why it has many dependencies on other libraries.

Install

It can be installed using pip:

pip3 install sent2vec

Usage

This is how to initialize the library and provide the data.

from sent2vec.vectorizer import Vectorizer

sentences = [
    "This is an awesome book to learn NLP.",
    "DistilBERT is an amazing NLP model.",
    "We can interchangeably use embedding, encoding, or vectorizing.",
]
vectorizer = Vectorizer(sentences)

If you want to use the the BERT language model to compute sentence embedding, you must use the code below.

vectors = vectorizer.bert()

Having the corresponding vectors, you can compute distance among vectors. Here, as expected, the distance between vectors[0] and vectors[1] is less than the distance between vectors[0] and vectors[2].

If you want to use a word2vec approach instead, you must first split sentences to lists of words using the sent2words method. In this stage, you can customized the list of stop-words by adding or removing to/from the default list. When you extract the most important words in sentences, you can compute the sentence embeddings using the w2v method. This method computes the average of vectors corresponding to the remaining words using the code bleow.

sentences = [
    "Alice is in the Wonderland.",
    "Alice is not in the Wonderland.",
]
vectorizer = Vectorizer(sentences)
vectorizer.sent2words(remove_stop_words=['not'], add_stop_words=[])
vectors = vectorizer.w2v()

And, that's pretty much it!