                        Changes in version 0.6.5                        

  - Change how word and document vectors are normalized when normalized
    = TRUE.
  - Export is_word2vec(), is_doc2vec() and check_model() for package
    development.
  - Improve the consistency of textmodel_word2vec and textmodel_doc2vec
    objects.
  - Add as.textmodel_doc2vec.tokens() with compound = TRUE to handle
    multi-word expressions more easily.

                 Changes in version 0.6.4 (2026-09-10)                  

  - Add as.textmodel_word2vec() to convert a matrix to a
    textmodel_word2vec object.
  - Update as.textmodel_doc2vec() to support conversion from the
    distributed memory (DM) model.
  - Add generic methods for probability() and similarity().
  - Change wordvector_threads to wordvector.threads in options with
    backward compatibility.

                 Changes in version 0.6.3 (2026-07-28)                  

  - Add padding to as.matrix() for using word2vec as a pre-trained model
    in torch.
  - Update internal functions for quanteda v4.5.0.

                 Changes in version 0.6.2 (2026-04-06)                  

  - Add layer to perplexity() for textmodel_doc2vec models.
  - Save document lengths as ntoken in trained textmodel_doc2vec models.
  - Update as.textmode_doc2vec() to save output layer weights.
  - Update tests for quanteda v4.4.0.

                 Changes in version 0.6.1 (2026-02-25)                  

  - Mention doc2vec in package description.
  - Add perplexity() to asses models' the goodness-of-fit to data.
  - Save quanteda's internal docvars in the textmodel_doc2vec objects.
  - Add group to as.matrix() to average sentence or paragraph vectors
    from the same documents.

                 Changes in version 0.6.0 (2025-12-09)                  

  - Upgrade textmodel_doc2vec to train the distributed memory (DM) and
    distributed bag-of-word (DBOW) models.
  - Add as.textmodel_doc2vec() to create document vectors as weighted
    average of word vectors.
  - Add layer to as.matrix() to choose between word or document vectors.
  - normalize is now defunct in textmodel_word2vec().

                 Changes in version 0.5.1 (2025-06-20)                  

  - Add normalize to textmodel_doc2vec() and pass it to as.matrix().
  - Add weights to textmodel_doc2vec() to adjust the salience of words
    in the document vectors.
  - Add include_data to textmodel_word2vec() to save the original tokens
    object.

                 Changes in version 0.5.0 (2025-05-15)                  

  - Add the model argument to textmodel_word2vec() to update existing
    models.
  - The normalize argument is moved from textmodel_word2vec() to
    as.matrix(). The original argument is deprecated and set to FALSE by
    default.
  - Remove weights().
  - Improve the structure of C++ code.

                        Changes in version 0.4.0                        

  - Add the tolower argument and set to TRUE to lower-case tokens.
  - Allow x to be quanteda's tokens_xptr object to enhance efficiency.

                 Changes in version 0.3.0 (2025-03-12)                  

  - Save docvars in the textmodel_doc2vec objects.
  - Set zero for empty documents in the textmodel_doc2vec objects.
  - Add probability() to compute probability of words.

                 Changes in version 0.2.0 (2025-01-07)                  

  - Rename word2vec(), doc2vec() and lsa() to textmodel_word2vec(),
    textmodel_doc2vec() and textmodel_lsa() respectively.
  - Simplify the C++ code to make maintenance easier.
  - Add normalize to word2vec to disable or enable word vector
    normalization.
  - Add weights() to extract back-propagation weights.
  - Make analogy() to convert a formula to named character vector.
  - Improve the stability of word2vec() when verbose = TRUE.

                 Changes in version 0.1.0 (2024-12-11)                  

  - Fork https://github.com/bnosac/word2vec and change the package name
    to wordvector.
  - Replace a list of character with quanteda's tokens object as an
    input object.
  - Recreate word2vec() with new argument names and object structures.
  - Create lda() to train word vectors using Latent Semantic Analysis.
  - Add similarity() and analogy() functions using proxyC.
  - Add data_corpus_news2014 that contain 20,000 news summaries as
    package data.
