Keyword extraction using transformer-based language models
Project description
WordWise
| | / /___ _________/ / | / (_)_______
| | /| / / __ \/ ___/ __ /| | /| / / / ___/ _ \
| |/ |/ / /_/ / / / /_/ / | |/ |/ / (__ ) __/
|__/|__/\____/_/ \__,_/ |__/|__/_/____/\___/
WordWise is a minimal keyword extraction library that uses quality contextual embeddings generated by Sentence-BERT to extract keywords from blocks of text.
Installation
WordWise is available on PyPI.
pip install wordwise
To clone this repository, run
git clone https://github.com/jaketae/wordwise.git
Quickstart
At the core of WordWise is the Extractor
class, which can be configured to generate keywords from some given text. The Extractor
can be initialized and used out-of-the-box with minimal configuration as follows.
from wordwise import Extractor
extractor = Extractor()
keywords = extractor.generate(text)
For advanced users, the Extractor
class, as well as the .generate()
method, can be configured in a more granular fashion to induce specific behaviors. For instance, the underlying spaCy backend can be specified.
extractor = Extractor(spacy_model="en_core_web_lg")
By default, the Extractor
will only generate the top 5 most relevant keywords. This behavior can be changed as follows.
# generate 10 keywords
keywords = extractor.generate(text, top_k=10)
Demo
Below is an example text adapted from the Wikipedia article on supervised learning.
text = """
Supervised learning is the machine learning task of
learning a function that maps an input to an output based
on example input-output pairs.[1] It infers a function
from labeled training data consisting of a set of
training examples.[2] In supervised learning, each
example is a pair consisting of an input object
(typically a vector) and a desired output value (also
called the supervisory signal). A supervised learning
algorithm analyzes the training data and produces an
inferred function, which can be used for mapping new
examples. An optimal scenario will allow for the algorithm
to correctly determine the class labels for unseen
instances. This requires the learning algorithm to
generalize from the training data to unseen situations
in a 'reasonable' way (see inductive bias).
"""
The extractor selects the three most relevant keywords from the block of text.
>>> from wordwise import Extractor
>>> extractor = Extractor()
>>> keywords = extractor.generate(text, 3)
>>> print(keywords)
['algorithm', 'learning', 'supervised learning']
How it Works
Using spaCy, the Extractor
object generates n-gram candidate noun phrases from the provided block of text. By default, it only considers uni-grams or bi-grams, since only rarely are keywords go beyond three words. Then, using a BERT model, it generates contextual embeddings for both the provided text and the n-gram keywords. Using cosine similarity as a distance function, it extracts the top_k
candidate keywords that are most similar to the embedding of the inputted text.
For a more detailed write-up, please refer to my blog post here.
Credit
WordWise was largely inspired by KeyBERT, a library that similarly uses sentence embeddings for keyword extraction. WordWise also relies on NLP libraries, such as spaCy and HuggingFace transformers, without which its development would not have been possible.
License
Released under the MIT License.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file wordwise-0.0.5.tar.gz
.
File metadata
- Download URL: wordwise-0.0.5.tar.gz
- Upload date:
- Size: 5.0 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.7.9
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | d24970ba2aa7d70a490b91e1da083eea3700c6902473e5c23654313980f77f25 |
|
MD5 | ba1e60ff180afcbb3f2f899921c8756e |
|
BLAKE2b-256 | ab84f2ba805f3f4505435327f9a632ec35a4c342bfb50b7a0a84311113e2cde1 |
File details
Details for the file wordwise-0.0.5-py3-none-any.whl
.
File metadata
- Download URL: wordwise-0.0.5-py3-none-any.whl
- Upload date:
- Size: 6.2 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.7.9
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 0d424b4b4e140c284186427494da21f79af533281ce0d2f513828821580e7ba8 |
|
MD5 | 8269f5256fa2644d114f067695459fde |
|
BLAKE2b-256 | 72d3f0adcc85c012940ddc06427389563735c6fec69ef4b8cacc087d4486b9c8 |