Building efficient recommendation systems in Ruby on Rails requires a thoughtful approach to both algorithm design and implementation. I’ve spent years working with recommendation engines, and I’ll share the most effective techniques I’ve discovered.

Collaborative Filtering Implementation

Collaborative filtering forms the foundation of modern recommendation systems. In Rails, we can implement user-based collaborative filtering effectively:

class CollaborativeFilter
  def initialize(user)
    @user = user
    @similarity_matrix = {}
  end

  def recommend
    calculate_user_similarities
    generate_recommendations
  end

  private

  def calculate_user_similarities
    User.find_each do |other_user|
      next if other_user.id == @user.id
      @similarity_matrix[other_user.id] = calculate_similarity(@user, other_user)
    end
  end

  def calculate_similarity(user1, user2)
    common_items = user1.ratings & user2.ratings
    return 0 if common_items.empty?
    
    pearson_correlation(user1.ratings, user2.ratings, common_items)
  end

  def generate_recommendations
    similar_users = @similarity_matrix.sort_by { |_, v| -v }.take(10)
    
    Item.find_by_sql([<<-SQL, @user.id, similar_users.map(&:first)])
      SELECT items.*, AVG(ratings.score) as predicted_rating
      FROM items
      JOIN ratings ON ratings.item_id = items.id
      WHERE ratings.user_id IN (?)
      AND items.id NOT IN (
        SELECT item_id FROM ratings WHERE user_id = ?
      )
      GROUP BY items.id
      ORDER BY predicted_rating DESC
      LIMIT 20
    SQL
  end
end

Content-Based Recommendations

Content-based filtering analyzes item attributes to make recommendations. Here’s my implementation using TF-IDF:

class ContentBasedRecommender
  def initialize
    @tfidf = TfIdfVectorizer.new
  end

  def process_items
    Item.find_each do |item|
      vector = @tfidf.transform(item.description)
      item.update(feature_vector: vector)
    end
  end

  def recommend_for_user(user)
    user_profile = calculate_user_profile(user)
    
    Item.find_each.sort_by do |item|
      cosine_similarity(user_profile, item.feature_vector)
    end.reverse.take(10)
  end

  private

  def calculate_user_profile(user)
    user.liked_items.map(&:feature_vector).reduce(&:+)
  end

  def cosine_similarity(v1, v2)
    dot_product = v1.zip(v2).map { |x, y| x * y }.sum
    magnitude = Math.sqrt(v1.map { |x| x**2 }.sum) * Math.sqrt(v2.map { |x| x**2 }.sum)
    dot_product / magnitude
  end
end

Performance Optimization

Caching plays a crucial role in recommendation system performance. I’ve developed this caching strategy:

class RecommendationCache
  def self.fetch_recommendations(user)
    Rails.cache.fetch("user_recommendations/#{user.id}", expires_in: 1.hour) do
      recommendations = generate_fresh_recommendations(user)
      update_recommendation_stats(user, recommendations)
      recommendations
    end
  end

  def self.generate_fresh_recommendations(user)
    Recommendation.transaction do
      collaborative_recommendations = CollaborativeFilter.new(user).recommend
      content_recommendations = ContentBasedRecommender.new.recommend_for_user(user)
      
      merge_recommendations(collaborative_recommendations, content_recommendations)
    end
  end

  private

  def self.merge_recommendations(collaborative, content)
    weighted_merge(collaborative, content, collaborative_weight: 0.7, content_weight: 0.3)
  end
end

User Behavior Tracking

Tracking user behavior provides valuable data for recommendations:

class UserBehaviorTracker
  include Sidekiq::Worker

  def perform(user_id, action, item_id)
    user = User.find(user_id)
    
    UserAction.create!(
      user: user,
      action_type: action,
      item_id: item_id,
      context: capture_context
    )

    update_user_preferences(user, action, item_id)
  end

  private

  def capture_context
    {
      timestamp: Time.current,
      session_duration: calculate_session_duration,
      device_type: detect_device_type,
      location: detect_location
    }
  end

  def update_user_preferences(user, action, item_id)
    preference = user.preferences.find_or_initialize_by(item_id: item_id)
    preference.score = calculate_preference_score(action)
    preference.save!
  end
end

A/B Testing Framework

Testing different recommendation algorithms requires a robust A/B testing framework:

class RecommendationExperiment
  def initialize(user)
    @user = user
    @experiment = create_experiment
  end

  def get_recommendations
    variant = @experiment.variant
    
    case variant
    when 'control'
      CollaborativeFilter.new(@user).recommend
    when 'treatment_a'
      ContentBasedRecommender.new.recommend_for_user(@user)
    when 'treatment_b'
      HybridRecommender.new(@user).recommend
    end
  end

  private

  def create_experiment
    Experiment.create!(
      name: 'recommendation_algorithm',
      user: @user,
      variants: ['control', 'treatment_a', 'treatment_b'],
      weights: [0.33, 0.33, 0.34]
    )
  end
end

Machine Learning Integration

Integrating machine learning enhances recommendation accuracy:

class MLRecommender
  def initialize
    @model = load_trained_model
  end

  def predict_ratings(user, items)
    features = extract_features(user, items)
    
    predictions = @model.predict(features)
    
    items.zip(predictions).sort_by { |_, rating| -rating }
  end

  private

  def extract_features(user, items)
    items.map do |item|
      [
        user.age,
        user.gender_encoded,
        item.category_encoded,
        item.price,
        user.average_rating,
        item.average_rating,
        interaction_count(user, item.category)
      ]
    end
  end

  def load_trained_model
    model_path = Rails.root.join('lib/models/recommendation_model.pkl')
    PyCall.import_module('joblib').load(model_path)
  end
end

Data Preprocessing

Proper data preprocessing ensures quality recommendations:

class DataPreprocessor
  def preprocess_user_data
    User.find_each do |user|
      normalized_ratings = normalize_ratings(user.ratings)
      user_features = extract_user_features(user)
      
      user.update!(
        normalized_ratings: normalized_ratings,
        feature_vector: user_features
      )
    end
  end

  def preprocess_item_data
    Item.find_each do |item|
      item_features = extract_item_features(item)
      popularity_score = calculate_popularity(item)
      
      item.update!(
        feature_vector: item_features,
        popularity_score: popularity_score
      )
    end
  end

  private

  def normalize_ratings(ratings)
    mean = ratings.average(:score)
    std = ratings.standard_deviation(:score)
    
    ratings.map { |r| (r.score - mean) / std }
  end

  def calculate_popularity(item)
    views = item.view_count
    ratings = item.ratings.count
    revenue = item.purchase_total
    
    (views * 0.3) + (ratings * 0.4) + (revenue * 0.3)
  end
end

This comprehensive approach to building recommendation systems in Rails provides a solid foundation for creating personalized user experiences. The combination of different techniques allows for flexible and accurate recommendations while maintaining good performance through proper caching and optimization strategies.