Latent Semantic Analysis (LSA) package

In brief

This Raku package, "ML::LatentSemanticAnalyzer", has different functions for computations of Latent Semantic Analysis (LSA) workflows (using Sparse matrix Linear Algebra.) The package mirrors the Mathematica implementation [AAp1]. (There is also a corresponding implementations in Python and R; see [AAp2, AAp3].)

The package provides:

Class ML::LatentSemanticAnalyzer
Functions for applying Latent Semantic Indexing (LSI) functions on matrix entries
"Data loader" function for obtaining a dataset of ~580 abstracts of conference presentations

Installation

To install from Zef ecosystem:

zef install ML::LatentSemanticAnalyzer

To install from GitHub use the shell command:

zef install https://github.com/antononcube/Raku-ML-LatentSemanticAnalyzer.git

LSA workflows

The scope of the package is to facilitate the creation and execution of the workflows encompassed in this flow chart:

LSA-workflows

For more details see the article "A monad for Latent Semantic Analysis workflows", [AA1].

Usage example

Here is an example of a LSA pipeline that:

Ingests a collection of texts
Makes the corresponding document-term matrix using stemming and removing stop words
Extracts 40 topics
Shows a table with the extracted topics
Shows a table with statistical thesaurus entries for selected words

use ML::LatentSemanticAnalyzer;
use Lingua::EN::Stem::Porter;

# Collection of texts
my @dsAbstracts = get-abstracts-dataset();
my %docs = @dsAbstracts.map(*<ID>) Z=> @dsAbstracts.map(*<Abstract>);
say %docs.elems;

# Remove non-strings
my %docs2 = %docs.grep({ $_.value ~~ Str:D });
say %docs2.elems;

# Stemmer object (to preprocess words in the pipeline below)
say &porter.WHY;

# Words to show statistical thesaurus entries for
my @words = <notebook computational function neural talk programming>;

# Reproducible results (just within a session)
srand(12);

# LSA pipeline
my $lsaObj =
        LatentSemanticAnalyzer.new
        .make-document-term-matrix(docs => %docs2, :stop-words, :stemming-rules, :3min-length)
        .apply_term_weight_functions( 
                global-weight-func => "IDF",
                local-weight-func => "None",
                normalizer-func => "Cosine")
        .extract-topics(:40number-of-topics, :10min-number-of-documents-per-term, method => "SVD")
        .echo-topics-interpretation(:12number-of-terms, :!wide-form)
        .echo_statistical_thesaurus(
                terms => @words.map(*.&porter),
                :wide-form,
                :12number-of-nearest-neighbors,
                method => "cosine",
                :echo)

This package is based on the Raku package "Math::SparseMatrix", [AAp5]

The package "ML::SparseMatrixRecommender", [AAp6] also uses LSI functions -- this package uses LSI methods of the class ML::SparseMatrixRecommender.

Mathematica

The Raku pipeline above corresponds to the following pipeline for the Mathematica package [AAp1]:

lsaObj =
  LSAMonUnit[aAbstracts]⟹
   LSAMonMakeDocumentTermMatrix["StemmingRules" -> Automatic, "StopWords" -> Automatic]⟹
   LSAMonEchoDocumentTermMatrixStatistics["LogBase" -> 10]⟹
   LSAMonApplyTermWeightFunctions["IDF", "None", "Cosine"]⟹
   LSAMonExtractTopics["NumberOfTopics" -> 20, Method -> "NNMF", "MaxSteps" -> 16, "MinNumberOfDocumentsPerTerm" -> 20]⟹
   LSAMonEchoTopicsTable["NumberOfTerms" -> 10]⟹
   LSAMonEchoStatisticalThesaurus["Words" -> Map[WordData[#, "PorterStem"]&, {"notebook", "computational", "function", "neural", "talk", "programming"}]];

Python

Here is a corresponding Python pipeline with the package [AAp3]:

lsaObj = (LatentSemanticAnalyzer()
          .make_document_term_matrix(docs=docs2,
                                     stop_words=True,
                                     stemming_rules=True,
                                     min_length=3)
          .apply_term_weight_functions(global_weight_func="IDF",
                                       local_weight_func="None",
                                       normalizer_func="Cosine")
          .extract_topics(number_of_topics=40, min_number_of_documents_per_term=10, method="NNMF")
          .echo_topics_interpretation(number_of_terms=12, wide_form=True)
          .echo_statistical_thesaurus(terms=stemmerObj.stemWords(words),
                                      wide_form=True,
                                      number_of_nearest_neighbors=12,
                                      method="cosine",
                                      echo_function=lambda x: print(x.to_string())))

R

The package LSAMon-R, [AAp2], implements a software monad for LSA workflows.

LSA packages comparison project

The project "Random mandalas deconstruction with R, Python, and Mathematica", [AAr1, AA2], has documents, diagrams, and (code) notebooks for comparison of LSA application to a collection of images (in multiple programming languages.)

A big part of the motivation to make the Python package "RandomMandala", [AAp4], was to make easier the LSA package comparison. Mathematica and R have fairly streamlined connections to Python, hence it is easier to propagate (image) data generated in Python into those systems.

Code generation with natural language commands

Using grammar-based interpreters

The project "Raku for Prediction", [AAr2, AAv2, AAp8], has a Domain Specific Language (DSL) grammar and interpreters that allow the generation of LSA code for corresponding Mathematica, Python, R, and Raku packages.

Here is Command Line Interface (CLI) invocation example that generate code for this package:

> ToLatentSemanticAnalysisWorkflowCode Python 'create from aDocs; apply LSI functions IDF, None, Cosine; extract 20 topics; show topics table'
# LatentSemanticAnalyzer(aDocs).apply_term_weight_functions(global_weight_func = "IDF", local_weight_func = "None", normalizer_func = "Cosine").extract_topics(number_of_topics = 20).echo_topics_table( )

NLP Template Engine

Here is an example using the NLP Template Engine, [AAr2, AAv3, AAp9]:

concretize("create from aDocs; apply LSI functions IDF, None, Cosine; extract 20 topics; show topics table", "TargetLanguage" -> "Raku")
(* 
my $lsaObj = LatentSemanticAnalyzer.new.
          .make-document-term-matrix(docs => aDocs, stop-words => Whatever, stemming-rules = Whatever, min-length => 3)
          .apply-term-weight-functions(global-weight-func => 'IDF', local-weight-func => 'None',normalizer-func => 'Cosine')
          .extract-topics(number-of-topics => 20, min-number-of-documents-per-term => 20, method => 'SVD')
          .echo-topics-interpretation(number-of-terms => 10, :wide-form)
          .echo-statistical-thesaurus(terms=["topics table"], :wide-form, number-of-nearest-neighbors => 12, method => 'cosine')
*)

Implementation details

The initial version was translated to Raku from the Python package "LatentSemanticAnalyzer", [AAp3].
- See "prompts.txt".
Multiple changes of the initial translation had to be made:
- Proper use of the (fast) Math::SparseMatrix::NativeAdapter matrices
- Proper topic extraction implementation (using SVD)
- Correct statistical thesaurus shaping
- Using "ML::SparseMatrixRecommender" for statistical thesaurus with Cosine distance
- Refactoring by moving multiple subs to ML::LatentSemanticAnalyzer::Utilities
- Correct (simplistic) resources data ingestion
- Correct automatic topic names derivation
- Profiling and re-implementation of impose-row-names and impose-column-names in "Math::SparseMatrix"
- Fixing a bug "Math::SparseMatrix" related to LSI application using a global weights list
- (Other changes...)
Testing against Python and Mathematica implementations with the same data and parameters.

References

Latent Semantic Analysis (LSA) package

In brief

Installation

LSA workflows

Usage example

Related Raku packages

Related Mathematica, Python, and R packages

Mathematica

Python

R

LSA packages comparison project

Code generation with natural language commands

Using grammar-based interpreters

NLP Template Engine

Implementation details

References

Articles

Python, R, and Wolfram Language packages

Raku packages

Repositories

Videos