Module: RetrievalLite::TfIdfRetrieval

Defined in:
lib/retrieval_lite/tfidf_retrieval.rb

Overview

See Also:

Class Method Summary collapse

Class Method Details

.evaluate(corpus, query) ⇒ Array<Document>

Queries a corpus using the tf-idf ranking algorithm and cosine similarity. Returns documents ordered by tf-idf score.

Parameters:

  • corpus (Corpus) —

    the collection of documents

  • query (String) —

    the boolean query to be evaluated

Returns:

  • (Array<Document>) —

    ordered array of documents that satisfy the query



9
10
11
 
# File 'lib/retrieval_lite/tfidf_retrieval.rb', line 9

def self.evaluate(corpus, query)
  evaluate_with_scores(corpus, query).keys
end

.evaluate_with_scores(corpus, query) ⇒ Hash<Document, Integer>

Queries a corpus using the tf-idf ranking algorithm and cosine similarity. Same as #evaluate but returns a hash whose keys are documents and values are the tf-idf score.

Parameters:

  • corpus (Corpus) —

    the collection of documents

  • query (String) —

    the boolean query to be evaluated

Returns:

  • (Hash<Document, Integer>) —

    ordered array of documents that satisfy the query



20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
 
# File 'lib/retrieval_lite/tfidf_retrieval.rb', line 20

def self.evaluate_with_scores(corpus, query)
  query_document = RetrievalLite::Document.new(query)
  terms = query_document.term_frequencies.keys
  query_vector = query_document.term_frequencies.values # should be in same order as keys

  documents = Set.new # ordering of documents doesn't matter right now
  # gathering only the documents that contain at least one of those terms
  terms.each do |t|
    docs_with_term = corpus.documents_with(t)
    if docs_with_term
      docs_with_term.each do |d|
        if !documents.include?(d)
          documents << d
        end
      end
    end
  end

  scores = {}
  documents.each do |document|
    document_vector = Array.new(terms.size)
    terms.each_with_index do |term, index|
      document_vector[index] = tfidf_weight(corpus, document, term)
    end
    scores[document] = RetrievalLite::Vector.cosine_similarity(query_vector, document_vector)
  end

  # order it by score in descending order
  return Hash[scores.sort_by{|key, value| value}.reverse]
end

.normalized_tfidf_weight(corpus, document, term) ⇒ Float

Ranks a document in corpus using the normalized tf-idf scoring.

Parameters:

Returns:

  • (Float) —

    the normalized tfidf weight of the term in the document

See Also:

  • #tfidf_weight


74
75
76
77
78
79
80
81
82
83
 
# File 'lib/retrieval_lite/tfidf_retrieval.rb', line 74

def self.normalized_tfidf_weight(corpus, document, term)
  length_of_vector = 0

  corpus.documents_with(term).each do |d|
    weight = tfidf_weight(corpus, d, term)
    length_of_vector += weight * weight
  end

  tfidf_weight(corpus, document, term) / Math.sqrt(length_of_vector)
end

.tfidf_weight(corpus, document, term) ⇒ Float

Note:

tf-idf is slightly modified. n_j (# of docs containing term j) is replaced with n_j + 1 to avoid divide by zero

Ranks a document in corpus using the tf-idf scoring.

Parameters:

Returns:

  • (Float) —

    the tfidf weight of the term in the document



59
60
61
62
63
64
65
 
# File 'lib/retrieval_lite/tfidf_retrieval.rb', line 59

def self.tfidf_weight(corpus, document, term)
  if corpus.document_frequency(term) == 0
    return 0
  else
    return document.frequency_of(term) * Math.log(1.0 * corpus.size/(corpus.document_frequency(term)))
  end
end