Class: Coelacanth::Extractor::MorphologicalAnalyzer

Inherits:
Object
  • Object
show all
Defined in:
lib/coelacanth/extractor/morphological_analyzer.rb

Overview

Scores candidate morphemes extracted from article content. The implementation follows a light-weight heuristic approach that approximates the specification shared in the user instructions. It prioritises noun-ish phrases for both Japanese and English text, applies positional boosts, and returns the highest scoring terms.

Defined Under Namespace

Classes: Term

Constant Summary collapse

TOKEN_PATTERN = 
/
  \p{Han}+ |                     # Kanji sequences
  \p{Hiragana}+ |                # Hiragana sequences
  [\p{Katakana}ー]+ |             # Katakana sequences including the choonpu
  [A-Za-z0-9]+(?:-[A-Za-z0-9]+)*  # Latin alphanumerics keeping inner hyphen
/x.freeze
MARKDOWN_CONTROL_PATTERN = 
/[`*_>#\[\]\(\)\{\}!\+=|~]/.freeze
EN_STOPWORDS = 
%w[
  a an and are as at be but by for if in into is it its of on or such
  that the their then there these they this to was were will with
].freeze
EN_GENERIC_TERMS = 
%w[
  page pages article articles category categories tag tags image images
  video videos click home link links read more author authors post posts
].freeze
JA_GENERIC_TERMS = 
%w[カテゴリ カテゴリー 記事 画像 写真 まとめ サイト 投稿 ブログ 最新 人気 関連].freeze
FULLWIDTH_ALPHA = 
"ＡＢＣＤＥＦＧＨＩＪＫＬＭＮＯＰＱＲＳＴＵＶＷＸＹＺａｂｃｄｅｆｇｈｉｊｋｌｍｎｏｐｑｒｓｔｕｖｗｘｙｚ".freeze
HALF_WIDTH_ALPHA = 
"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz".freeze
FULLWIDTH_DIGITS = 
"０１２３４５６７８９".freeze
HALF_WIDTH_DIGITS = 
"0123456789".freeze
FULLWIDTH_HYPHENS = 
"－―ーｰ".freeze
TOP_K = 
8
MAX_POSITION_BOOST = 
3.0
LENGTH_BONUS_FACTOR = 
0.15
MAX_LENGTH_BONUS = 
1.6
POSITION_WEIGHTS = 
{
  body: 1.0,
  title: 2.2,
  h1: 1.6,
  h2: 1.3,
  accent: 1.1
}.freeze
CATEGORY_ALIASES = 
{
  "kanji" => :kanji,
  "hiragana" => :hiragana,
  "katakana" => :katakana,
  "latin" => :latin
}.freeze

Instance Method Summary collapse

Constructor Details

#initialize(config: Coelacanth.config) ⇒ MorphologicalAnalyzer

Returns a new instance of MorphologicalAnalyzer.



63
64
65
 
# File 'lib/coelacanth/extractor/morphological_analyzer.rb', line 63

def initialize(config: Coelacanth.config)
  @config = config
end

Instance Method Details

#call(node:, title:, markdown:) ⇒ Object 



71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
 
# File 'lib/coelacanth/extractor/morphological_analyzer.rb', line 71

def call(node:, title:, markdown:)
  stats = Hash.new do |hash, key|
    hash[key] = {
      token: nil,
      components: 1,
      body_count: 0,
      pos_bonus: 0.0,
      language: nil
    }
  end

  body_terms = extract_terms(markdown)

  contexts = [
    [POSITION_WEIGHTS[:title], extract_terms(title)],
    [POSITION_WEIGHTS[:h1], extract_terms(text_for(node, "h1"))],
    [POSITION_WEIGHTS[:h2], extract_terms(text_for(node, "h2"))],
    [
      POSITION_WEIGHTS[:accent],
      extract_terms(
        [
          text_for(node, "a"),
          text_for(node, "strong"),
          text_for(node, "b"),
          text_for(node, "img", attribute: "alt")
        ].compact.join(" ")
      )
    ],
    [POSITION_WEIGHTS[:body], body_terms]
  ]

  contexts.each do |weight, terms|
    next if terms.empty?

    grouped = terms.group_by(&:key)
    grouped.each_value do |occurrences|
      representative = occurrences.max_by(&:components)
      entry = stats[representative.key]
      entry[:token] ||= representative.token
      entry[:components] = [entry[:components], representative.components].max
      entry[:language] ||= representative.language

      bonus = weight - 1.0
      entry[:pos_bonus] += bonus if bonus.positive?
    end
  end

  body_terms.each do |term|
    entry = stats[term.key]
    entry[:token] ||= term.token
    entry[:components] = [entry[:components], term.components].max
    entry[:language] ||= term.language
    entry[:body_count] += 1
  end

  scored = stats.values.map do |entry|
    next if entry[:body_count].zero?

    tf = Math.log(entry[:body_count] + 1.0)
    pos_boost = [1.0 + entry[:pos_bonus], MAX_POSITION_BOOST].min
    len_bonus = [1.0 + LENGTH_BONUS_FACTOR * (entry[:components] - 1), MAX_LENGTH_BONUS].min
    score = tf * pos_boost * len_bonus

    entry.merge(score: score)
  end.compact

  return [] if scored.empty?

  sorted = scored.sort_by { |entry| [-entry[:score], entry[:token]] }
  pruned = prune_inclusions(sorted)
  max_score = pruned.first[:score]
  threshold = max_score * 0.35

  selected = pruned.select { |entry| entry[:score] >= threshold }

  if selected.length < TOP_K
    pruned.each do |entry|
      next if selected.include?(entry)

      selected << entry
      break if selected.length >= TOP_K
    end
  end

  selected = selected.take(TOP_K)

  selected.map do |entry|
    {
      token: entry[:token],
      score: entry[:score],
      count: entry[:body_count]
    }
  end
end

#call_text(text, title: nil) ⇒ Object 



67
68
69
 
# File 'lib/coelacanth/extractor/morphological_analyzer.rb', line 67

def call_text(text, title: nil)
  call(node: nil, title: title, markdown: text)
end