Detokenization now restores em/en, third/quarter, thin/hair, medium math space characters & narrow non breaking space characters besides the regular nbsp. 👏 🙌 🛰️

2.2.2

published 2.2.2 • 7 months ago

Changelog

.contextualVectors() now throws error if (a) word vectors are not loaded and (b) with lemma: true, "pos" is missing in the NLP pipe. 🤓

2.2.1

published 2.2.1 • 7 months ago

Changelog

Added missing typescript definitions for word embeddings besides few other typescript fixes. ✅

published 2.2.0 • 9 months ago

Changelog

Detokenization restores both regular and non-breaking spaces to their original positions. 🤓

published 2.1.0 • 9 months ago

Changelog

You can now use similarity.vector.cosine( vectorA, vectorB ) to compute similarity between two vectors on a scale of 0 to 1. 🤓

published 2.0.0 • 9 months ago

Changelog

Seamless word embedding integration enhances winkNLP's semantic capabilities. 🎉 👏 🙌
Pre-trained 100-dimensional word embeddings for over 350,000 English words released: wink-embeddings-sg-100d. 💯
API remains unchanged — no code updates needed for existing projects. The new APIs include: 🤩
- Obtain vector for a token: Use the .vectorOf( token ) API.
- Compute sentence/document embeddings: Employ the as.vector helper: use .out( its.lemma, as.vector ) on tokens of a sentence or document. You can also use its.value or its.normal. Tokens can be pre-processed to remove stop words etc using the .filter() API. Note, the as.vector helper uses averaging technique.
- Generate contextual vectors: Leverage the .contextualVectors() method on a document. Useful for pure browser-side applications! Generate custom vectors contextually relevant to your corpus and use them in place of larger pre-trained wink embeddings.
Comprehensive documentation along with interesting examples is coming up shortly. Stay tuned for updates! 😎