Recommender Systems

  1. Content-Based Filtering
    1. Core Principles
      1. Matching User and Item Profiles
        1. Personalization Based on Item Features
          1. Independence from Other Users
          2. Item Profile Creation
            1. Feature Extraction from Items
              1. Textual Features
                1. Bag-of-Words
                  1. TF-IDF
                    1. Word Embeddings
                      1. N-grams
                      2. Categorical Features
                        1. Genre
                          1. Brand
                            1. Tags
                              1. Categories
                              2. Numerical Features
                                1. Price
                                  1. Release Year
                                    1. Duration
                                    2. Multimedia Features
                                      1. Visual Feature Extraction
                                        1. Audio Feature Extraction
                                      2. Item Representation
                                        1. Vector Space Models
                                          1. Feature Vectors
                                            1. Dimensionality Reduction
                                              1. Principal Component Analysis
                                                1. Singular Value Decomposition
                                            2. User Profile Creation
                                              1. Aggregating Profiles of Liked Items
                                                1. Summing Feature Vectors
                                                  1. Averaging Feature Vectors
                                                  2. Weighted Averages of Item Features
                                                    1. Incorporating User Ratings
                                                      1. Time-Weighted Aggregation
                                                      2. Learning User Preferences
                                                        1. Preference Elicitation
                                                          1. Profile Evolution
                                                        2. Generating Recommendations
                                                          1. Measuring Similarity
                                                            1. Cosine Similarity
                                                              1. Euclidean Distance
                                                                1. Jaccard Similarity
                                                                  1. Pearson Correlation
                                                                  2. Ranking Items by Similarity Score
                                                                    1. Top-N Recommendation
                                                                      1. Threshold-Based Filtering
                                                                      2. Score Normalization
                                                                      3. Advantages and Disadvantages
                                                                        1. Strengths
                                                                          1. User Independence
                                                                            1. Transparency and Explainability
                                                                              1. No Cold-Start for New Items
                                                                                1. Adaptability to User Preferences
                                                                                2. Weaknesses
                                                                                  1. Limited Serendipity
                                                                                    1. Requires Domain Knowledge for Feature Engineering
                                                                                      1. User Cold-Start Problem
                                                                                        1. Limited Discovery of Novel Items

                                                                                    Previous

                                                                                    2. Data Foundations and Preprocessing

                                                                                    Go to top

                                                                                    Next

                                                                                    4. Collaborative Filtering