sbert 0.6.0

This release collects the changes made since the CRAN release 0.5.8.

New features

Changes in behavior

Bug fixes

Documentation

Other changes

sbert 0.5.7

sbert 0.5.6

sbert 0.5.5

sbert 0.5.4

sbert 0.5.3

sbert 0.5.2

sbert 0.5.1

sbert 0.5.0

Breaking: the sbert_ prefix is gone

Every exported function lost its prefix. The package has never been released, so the old names were removed outright rather than deprecated:

sbert_topics() is now topics(), sbert_encode() encode(), sbert_segment() segment(), sbert_blend() blend(), sbert_keywords() keywords(), sbert_dedupe() dedupe(), sbert_coherence() coherence(), sbert_select_topics() select_topics(), sbert_representatives() representatives(), sbert_load_model() load_model(), sbert_models() models(), and so on for the cache and runtime verbs.

Six names are qualified because the bare form collides with a package this audience is likely to have attached — igraph, quanteda, purrr:

Two are qualified because the bare form would mask base R: sbert_gamma() -> topic_gamma(), sbert_palette() -> topic_palette().

Class names keep the prefix (sbert_topic_model, sbert_model), so S3 methods read terms.sbert_topic_model().

New: terms() retunes without refitting

Topic terms depend only on the fitted assignments and the document text, so every term setting can now be changed without repeating the clustering or the encoding:

terms(fit, n = 12, weighting = "bm25", stem = TRUE)
terms(fit, n = 8, sort_by = "beta")     # words a topic uses most
terms(fit, n = 8)                        # words that distinguish it

sort_by chooses between the class-based weight (distinctive terms, the default) and raw p(term | topic). n is applied after ordering. The returned frame carries beta, which absorbs the removed sbert_beta().

Verbs take your data frame and keep what they compute

select_topics() keeps every model it fits

The sweep no longer discards the models it built. fitted(sweep, n_topics =) returns one without refitting, and the result gains print, plot and as.data.frame methods while remaining a plain data frame with the same columns.

Other

sbert 0.4.0

Static embeddings (instant-speed tier)

Seeded (guided) topics

Analysis utilities

Context-blended segment embeddings

Representative selection and segmentation

Multi-model registry

Generic loader

One-verb usage

Corpus verbs

Inferential layer

sbert 0.3.0

sbert 0.2.0

sbert 0.1.0