svs-package: Tools for Semantic Vector Spaces
In svs: Tools for Semantic Vector Spaces

svs-package

R Documentation

Tools for Semantic Vector Spaces

Description

This package offers various tools for semantic vector spaces. There are techniques for correspondence analysis (simple, multiple and discriminant), latent semantic analysis, probabilistic latent semantic analysis, non-negative matrix factorization, latent class analysis, EM clustering, logratio analysis and log-multiplicative (association) analysis. Furthermore, the package has specialized distance measures and plotting functions as well as some helper functions.

This package contains the following raw data files (in the folder extdata):

SndT_Fra.txt: Seventeen Dutch source words and their French translations.
SndT_Eng.txt: Seventeen Dutch source words and their English translations.
InvT_Fra.txt: Seventeen Dutch target words and their French source words.
InvT_Eng.txt: Seventeen Dutch target words and their English source words.
Ctxt_Dut.txt: Context words for seventeen Dutch words.
Ctxt_Fra.txt: Context words for seventeen Dutch words translated from French.
Ctxt_Eng.txt: Context words for seventeen Dutch words translated from English.

The (fast procedures for the) techniques in this package are:

fast_sca: Simple correspondence analysis.
fast_mca: Multiple correspondence analysis.
fast_dca: Discriminant correspondence analysis.
fast_lsa: Latent semantic analysis.
fast_psa: Probabilistic latent semantic analysis.
fast_nmf: Non-negative matrix factorization.
fast_lca: Latent class analysis.
fast_E_M: EM clustering.
fast_lra: Logratio analysis.
fast_lma: Log-multiplicative (association) analysis.

The complete overview of local and global weighting functions in this package can be found on weighting_functions.

The specialized distance measures are:

dist_chisquare: Chi-square distance.
dist_cosine: Cosine distance.
dist_wrt: Distance with respect to a certain point.
dist_wrt_centers: Distance with respect to cluster centers.

The specialized plotting functions are:

cd_plot: Cumulative distribution plot.
pc_plot: Parallel coordinate plot.

There are two helper functions for correspondence analysis:

freq_ca: Compute level frequencies (for a factor).
centers_ca: Compute coordinates for cluster centers.

There is one helper function for pvclust:

complete_pvpick: Complete the output of pvpick.

There is one helper function for igraph:

layout4bipartite: Create a layout matrix for a bipartite graph.

The remaining helper functions in this package are:

rep4dat: Repeat the rows of a data frame according to a frequency column.
vec2ddc: Transform a vector into a double-coded matrix.
dat2ddc: Transform a data frame into a double-coded matrix.
vec2ind: Transform a vector into an indicator matrix.
tab2dat: Transform a table into a data frame.
tab2ind: Transform a table into an indicator matrix.
dat2ind: Transform a data frame into an indicator matrix.
outerec: Recursive application of the outer product.
pmi: Pointwise mutual information.
MI: Mutual information.
log_or_0: Logarithmic transform.

Further reference

Many packages contain correspondence analysis: ca, FactoMineR, MASS and others.
For latent semantic analysis there is also the package lsa.
The package NMF provides more flexibility for non-negative matrix factorization.
For topic models there are the packages lda and topicmodels.
Latent class analysis can also be run in the package poLCA.
For log-ratio analysis there is also the package easyCODA.
The package gnm offers much flexibility for association analysis, i.e. fitting log-multiplicative or Goodman's RC models.

Link

As from 2023, this package is part of Module 10: Multivariate data analysis with R of the Summer School Methods in Language Sciences.

Author

Koen Plevoets, koen.plevoets@ugent.be

Acknowledgements

This package has benefited greatly from the helpful comments of Lore Vandevoorde, Pauline De Baets and Gert De Sutter. Thanks to Kurt Hornik, Uwe Ligges and Brian Ripley for their valuable recommendations when proofing this package.

svs documentation built on June 24, 2024, 5:07 p.m.