LoginSign Up
← Back to all sparks
D

doc2vec

AI-ASSISTANTS
Velocity0.0

Distributed Representations of Sentences, Documents and Topics

github.com/bnosac/doc2vec ↗

doc2vec's one directional release added topic discovery to a document-embedding package

nlpembeddingstopic-modelingr-packagebnosac
◆Current state
doc2vec wraps a C++ paragraph2vec implementation for R, training document and word embeddings from raw text. Its 0.2.0 release added the top2vec semantic clustering algorithm and support for initialising word embeddings from a pretrained set, which is where the package's current capability surface was set. Since then it has been quiet: the 2025 release only fixes a DOI in DESCRIPTION and drops the C++11 declaration from Makevars.
◆Where it's heading
This is a settled member of the bnosac NLP family and moves with it rather than on its own schedule. The same C++11 Makevars cleanup landed across word2vec and BTM within a day of this release, which is the shape of a CRAN compliance sweep over a maintainer's whole set rather than package-level development. Nothing in five years suggests further algorithm work is planned here.
◆Prediction
Expect the next release to be another cross-package compliance pass triggered by a CRAN or toolchain change, not new modelling capability.

◆Recent moves

  1. 8mo ago

    Fix DESCRIPTION DOI; drop C++11 from Makevars

    Two packaging corrections, both of which landed across the sibling bnosac packages within roughly a day. CRAN compliance rather than package development.

    View source ↗
  2. 5y ago

    top2vec clustering and transfer learning from pretrained vectors

    ⚡ SPARK

    Adds the top2vec semantic clustering algorithm and allows paragraph2vec to initialise its word embeddings from a pretrained set, plus a Belgian parliament question dataset and a file-handle fix. Everything the package can currently do beyond plain paragraph2vec dates from here.

    View source ↗
  3. 5y ago

    Valgrind fixes; WMD removed and udpipe suggestion dropped

    Fixes flagged by valgrind checks including Vocabulary destructors, removes WMD, and adds txt_count_words so the udpipe suggestion can be dropped. Cleanup shortly after the initial release.

    View source ↗
  4. 5y ago

    Initial release wrapping the hiyijian doc2vec implementation

    The first CRAN release, packaging an existing C++ doc2vec implementation for R. A straight binding, with the algorithmic additions arriving three months later in 0.2.0.

    View source ↗