Class: Reckon::CosineSimilarity

Inherits:
Object
  • Object
show all
Defined in:
lib/reckon/cosine_similarity.rb

Instance Method Summary collapse

Constructor Details

#initialize(options) ⇒ CosineSimilarity

Returns a new instance of CosineSimilarity.



8
9
10
11
12
 
# File 'lib/reckon/cosine_similarity.rb', line 8

def initialize(options)
  @options = options
  @tokens = {}
  @accounts = Hash.new(0)
end

Instance Method Details

#add_document(account, doc) ⇒ Object 



14
15
16
17
18
19
20
21
22
23
 
# File 'lib/reckon/cosine_similarity.rb', line 14

def add_document(, doc)
  tokenize(doc).each do |n|
    (token, count) = n

    @tokens[token] ||= {}
    @tokens[token][] ||= 0
    @tokens[token][] += count
    @accounts[] += count
  end
end

#find_similar(query) ⇒ Object

find most similar documents to query



26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
 
# File 'lib/reckon/cosine_similarity.rb', line 26

def find_similar(query)
  (query_scores, corpus_scores) = td_idf_scores_for(query)

  query_vector = Vector.elements(query_scores, false)

  # For each doc, calculate the similarity to the query
  suggestions = corpus_scores.map do |, scores|
    acct_vector = Vector.elements(scores, false)

    acct_query_dp = acct_vector.inner_product(query_vector)
    # similarity is a float between 1 and -1, where 1 is exactly the same and -1 is
    # exactly opposite
    # see https://en.wikipedia.org/wiki/Cosine_similarity
    # cos(theta) = (A . B) / (||A|| ||B||)
    # where A . B is the "dot product" and ||A|| is the magnitude of A
    # ruby has the 'matrix' library we can use to do these calculations.
    {
      similarity: acct_query_dp / (acct_vector.magnitude * query_vector.magnitude),
      account: ,
    }
  end.select { |n| n[:similarity] > 0 }.sort_by { |n| -n[:similarity] }

  LOGGER.info "most similar accounts: #{suggestions}"

  return suggestions
end