Module: Recommendable::Helpers::Calculations

Defined in:
lib/recommendable/helpers/calculations.rb

Class Method Summary collapse

Class Method Details

.predict_for(user_id, klass, item_id) ⇒ Float

Predict how likely it is that a user will like an item. This probability is not based on percentage. 0.0 indicates that the user will neither like nor dislike the item. Values that approach Infinity indicate a rising likelihood of liking the item while values approaching -Infinity indicate a rising probability of disliking the item.

Parameters:

  • user_id (Fixnum, String) —

    the user’s ID

  • klass (Class) —

    the item’s class

  • item_id (Fixnum, String) —

    the item’s ID

Returns:

  • (Float) —

    the probability that the user will like the item



165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
 
# File 'lib/recommendable/helpers/calculations.rb', line 165

def predict_for(user_id, klass, item_id)
  user_id = user_id.to_s
  item_id = item_id.to_s

  liked_by_set = Recommendable::Helpers::RedisKeyMapper.liked_by_set_for(klass, item_id)
  disliked_by_set = Recommendable::Helpers::RedisKeyMapper.disliked_by_set_for(klass, item_id)
  similarity_sum = 0.0

  similarity_sum += similarity_total_for(user_id, liked_by_set)
  similarity_sum -= similarity_total_for(user_id, disliked_by_set)

  liked_by_count, disliked_by_count = Recommendable.redis.pipelined do
    Recommendable.redis.scard(liked_by_set)
    Recommendable.redis.scard(disliked_by_set)
  end
  prediction = similarity_sum / (liked_by_count + disliked_by_count).to_f
  prediction.finite? ? prediction : 0.0
end

.similarity_between(user_id, other_user_id) ⇒ Float

Note:

Similarity values are asymmetrical. ‘Calculations.similarity_between(user_id, other_user_id)` will not necessarily equal `Calculations.similarity_between(other_user_id, user_id)`

Calculate a numeric similarity value that can fall between -1.0 and 1.0. A value of 1.0 indicates that both users have rated the same items in the same ways. A value of -1.0 indicates that both users have rated the same items in opposite ways.

Parameters:

  • user_id (Fixnum, String) —

    the ID of the first user

  • other_user_id (Fixnum, String) —

    the ID of another user

Returns:

  • (Float) —

    the numeric similarity between this user and the passed user



14
15
16
17
18
19
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
50
51
 
# File 'lib/recommendable/helpers/calculations.rb', line 14

def similarity_between(user_id, other_user_id)
  user_id = user_id.to_s
  other_user_id = other_user_id.to_s

  similarity = liked_count = disliked_count = 0
  in_common = Recommendable.config.ratable_classes.each do |klass|
    liked_set = Recommendable::Helpers::RedisKeyMapper.liked_set_for(klass, user_id)
    other_liked_set = Recommendable::Helpers::RedisKeyMapper.liked_set_for(klass, other_user_id)
    disliked_set = Recommendable::Helpers::RedisKeyMapper.disliked_set_for(klass, user_id)
    other_disliked_set = Recommendable::Helpers::RedisKeyMapper.disliked_set_for(klass, other_user_id)

    results = Recommendable.redis.pipelined do
      # Agreements
      Recommendable.redis.sinter(liked_set, other_liked_set)
      Recommendable.redis.sinter(disliked_set, other_disliked_set)

      # Disagreements
      Recommendable.redis.sinter(liked_set, other_disliked_set)
      Recommendable.redis.sinter(disliked_set, other_liked_set)

      Recommendable.redis.scard(liked_set)
      Recommendable.redis.scard(disliked_set)
    end

    # Agreements
    similarity += results[0].size
    similarity += results[1].size

    # Disagreements
    similarity -= results[2].size
    similarity -= results[3].size

    liked_count += results[4]
    disliked_count += results[5]
  end

  similarity / (liked_count + disliked_count).to_f
end

.similarity_total_for(user_id, set) ⇒ Object 



184
185
186
187
188
189
190
191
192
193
 
# File 'lib/recommendable/helpers/calculations.rb', line 184

def similarity_total_for(user_id, set)
  similarity_set = Recommendable::Helpers::RedisKeyMapper.similarity_set_for(user_id)
  ids = Recommendable.redis.smembers(set)
  similarity_values = Recommendable.redis.pipelined do
    ids.each do |id|
      Recommendable.redis.zscore(similarity_set, id)
    end
  end
  similarity_values.map(&:to_f).reduce(&:+).to_f
end

.update_recommendations_for(user_id) ⇒ Object

Used internally to update this user’s prediction values across all recommendable types. This is called by the background worker.



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
 
# File 'lib/recommendable/helpers/calculations.rb', line 103

def update_recommendations_for(user_id)
  user_id = user_id.to_s

  nearest_neighbors = Recommendable.config.nearest_neighbors || Recommendable.config.user_class.count
  Recommendable.config.ratable_classes.each do |klass|
    rated_sets = [
      Recommendable::Helpers::RedisKeyMapper.liked_set_for(klass, user_id),
      Recommendable::Helpers::RedisKeyMapper.disliked_set_for(klass, user_id),
      Recommendable::Helpers::RedisKeyMapper.hidden_set_for(klass, user_id),
      Recommendable::Helpers::RedisKeyMapper.bookmarked_set_for(klass, user_id)
    ]
    temp_set = Recommendable::Helpers::RedisKeyMapper.temp_set_for(Recommendable.config.user_class, user_id)
    similarity_set  = Recommendable::Helpers::RedisKeyMapper.similarity_set_for(user_id)
    recommended_set = Recommendable::Helpers::RedisKeyMapper.recommended_set_for(klass, user_id)
    most_similar_user_ids, least_similar_user_ids = Recommendable.redis.pipelined do
      Recommendable.redis.zrevrange(similarity_set, 0, nearest_neighbors - 1)
      Recommendable.redis.zrange(similarity_set, 0, nearest_neighbors - 1)
    end

    # Get likes from the most similar users
    sets_to_union = most_similar_user_ids.inject([]) do |sets, id|
      sets << Recommendable::Helpers::RedisKeyMapper.liked_set_for(klass, id)
    end

    # Get dislikes from the least similar users
    sets_to_union = least_similar_user_ids.inject(sets_to_union) do |sets, id|
      sets << Recommendable::Helpers::RedisKeyMapper.disliked_set_for(klass, id)
    end

    return if sets_to_union.empty?

    # SDIFF rated items so they aren't recommended
    Recommendable.redis.sunionstore(temp_set, *sets_to_union)
    item_ids = Recommendable.redis.sdiff(temp_set, *rated_sets)
    scores = item_ids.map { |id| [predict_for(user_id, klass, id), id] }
    Recommendable.redis.pipelined do
      scores.each do |s|
        Recommendable.redis.zadd(recommended_set, s[0], s[1])
      end
    end

    Recommendable.redis.del(temp_set)

    if number_recommendations = Recommendable.config.recommendations_to_store
      length = Recommendable.redis.zcard(recommended_set)
      Recommendable.redis.zremrangebyrank(recommended_set, 0, length - number_recommendations - 1)
    end
  end

  true
end

.update_score_for(klass, id) ⇒ Object 



195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
 
# File 'lib/recommendable/helpers/calculations.rb', line 195

def update_score_for(klass, id)
  score_set = Recommendable::Helpers::RedisKeyMapper.score_set_for(klass)
  liked_by_set = Recommendable::Helpers::RedisKeyMapper.liked_by_set_for(klass, id)
  disliked_by_set = Recommendable::Helpers::RedisKeyMapper.disliked_by_set_for(klass, id)
  liked_by_count, disliked_by_count = Recommendable.redis.pipelined do
    Recommendable.redis.scard(liked_by_set)
    Recommendable.redis.scard(disliked_by_set)
  end

  return 0.0 unless liked_by_count + disliked_by_count > 0

  z = 1.96
  n = liked_by_count + disliked_by_count
  phat = liked_by_count / n.to_f

  begin
    score = (phat + z*z/(2*n) - z * Math.sqrt((phat*(1-phat)+z*z/(4*n))/n))/(1+z*z/n)
  rescue Math::DomainError
    score = 0
  end

  Recommendable.redis.zadd(score_set, score, id)
  true
end

.update_similarities_for(user_id) ⇒ Object

Used internally to update the similarity values between this user and all other users. This is called by the background worker.



55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
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
 
# File 'lib/recommendable/helpers/calculations.rb', line 55

def update_similarities_for(user_id)
  user_id = user_id.to_s # For comparison. Redis returns all set members as strings.
  similarity_set = Recommendable::Helpers::RedisKeyMapper.similarity_set_for(user_id)

  # Only calculate similarities for users who have rated the items that
  # this user has rated
  relevant_user_ids = Recommendable.config.ratable_classes.inject([]) do |memo, klass|
    liked_set = Recommendable::Helpers::RedisKeyMapper.liked_set_for(klass, user_id)
    disliked_set = Recommendable::Helpers::RedisKeyMapper.disliked_set_for(klass, user_id)

    item_ids = Recommendable.redis.sunion(liked_set, disliked_set)

    unless item_ids.empty?
      sets = item_ids.map do |id|
        liked_by_set = Recommendable::Helpers::RedisKeyMapper.liked_by_set_for(klass, id)
        disliked_by_set = Recommendable::Helpers::RedisKeyMapper.disliked_by_set_for(klass, id)

        [liked_by_set, disliked_by_set]
      end

      memo | Recommendable.redis.sunion(*sets.flatten)
    else
      memo
    end
  end

  similarity_values = relevant_user_ids.map { |id| similarity_between(user_id, id) }
  Recommendable.redis.pipelined do
    relevant_user_ids.zip(similarity_values).each do |id, similarity_value|
      next if id == user_id # Skip comparing with self.
      Recommendable.redis.zadd(similarity_set, similarity_value, id)
    end
  end

  if knn = Recommendable.config.nearest_neighbors
    length = Recommendable.redis.zcard(similarity_set)
    kfn = Recommendable.config.furthest_neighbors || 0

    Recommendable.redis.zremrangebyrank(similarity_set, kfn, length - knn - 1)
  end

  true
end