\contentsline {schapter}{{List of illustrations}}{x}{chapter*.2}
\contentsline {schapter}{List of tables}{xii}{chapter*.3}
\contentsline {schapter}{Notation}{xiii}{chapter*.4}
\contentsline {chapter}{\numberline {1}Introduction}{1}{chapter.1}
\contentsline {section}{\numberline {1.1}Purpose of This Book}{2}{section.1.1}
\contentsline {section}{\numberline {1.2}For Learners: What is Covered, and What Isn't}{3}{section.1.2}
\contentsline {paragraph}{Regressors, classifiers, and the learning pipeline}{3}{section*.7}
\contentsline {paragraph}{User representations and dimensionality reduction}{3}{section*.8}
\contentsline {paragraph}{Deep learning}{3}{section*.9}
\contentsline {paragraph}{Offline versus online learning}{4}{section*.10}
\contentsline {paragraph}{Bias, consequences, and user considerations}{4}{section*.11}
\contentsline {paragraph}{Implementation and libraries}{4}{section*.12}
\contentsline {section}{\numberline {1.3}For Instructors: Course and Content Outline}{5}{section.1.3}
\contentsline {subsection}{\numberline {1.3.1}Course Plan and Overview}{6}{subsection.1.3.1}
\contentsline {paragraph}{Machine Learning Primer}{6}{section*.13}
\contentsline {paragraph}{Recommender Systems}{6}{section*.14}
\contentsline {paragraph}{Content and Structure in Recommender Systems}{6}{section*.15}
\contentsline {paragraph}{Temporal and Sequential Models}{7}{section*.16}
\contentsline {paragraph}{Personalized Models of Text}{7}{section*.17}
\contentsline {paragraph}{Personalized Models of Visual Data}{7}{section*.18}
\contentsline {paragraph}{The Consequences of Personalized Machine Learning}{7}{section*.19}
\contentsline {section}{\numberline {1.4}Online Resources}{7}{section.1.4}
\contentsline {section}{\numberline {1.5}About the Author}{8}{section.1.5}
\contentsline {section}{\numberline {1.6}Personalization in Everyday Life}{9}{section.1.6}
\contentsline {subsection}{\numberline {1.6.1}Recommendation}{9}{subsection.1.6.1}
\contentsline {subsection}{\numberline {1.6.2}Personalized Health}{10}{subsection.1.6.2}
\contentsline {subsection}{\numberline {1.6.3}Computational Social Science}{11}{subsection.1.6.3}
\contentsline {subsection}{\numberline {1.6.4}Language Generation, Personalized Dialog, and Interactive Agents}{11}{subsection.1.6.4}
\contentsline {section}{\numberline {1.7}Techniques for Personalization}{12}{section.1.7}
\contentsline {subsection}{\numberline {1.7.1}User Representations as Manifolds}{12}{subsection.1.7.1}
\contentsline {subsection}{\numberline {1.7.2}Contextual Personalization and Model-Based Personalization}{13}{subsection.1.7.2}
\contentsline {section}{\numberline {1.8}The Ethics and Consequences of Personalization}{14}{section.1.8}
\contentsline {part}{\MakeUppercase {Part\ ONE\relax \hskip 1em\relax Machine Learning Primer}}{17}{part.1}
\contentsline {chapter}{\numberline {2}Regression and Feature Engineering}{19}{chapter.2}
\contentsline {subsubsection}{Supervised learning}{20}{section*.20}
\contentsline {section}{\numberline {2.1}Linear Regression}{21}{section.2.1}
\contentsline {paragraph}{More complex models}{23}{section*.21}
\contentsline {paragraph}{Adding more dimensions}{24}{section*.22}
\contentsline {subsection}{\numberline {2.1.1}Regression in \emph {sklearn}}{24}{subsection.2.1.1}
\contentsline {section}{\numberline {2.2}Evaluating Regression Models}{26}{section.2.2}
\contentsline {subsection}{\numberline {2.2.1}The Mean Squared Error}{26}{subsection.2.2.1}
\contentsline {subsection}{\numberline {2.2.2}Why the Mean Squared Error?}{26}{subsection.2.2.2}
\contentsline {subsection}{\numberline {2.2.3}Maximum Likelihood Estimation of Model Parameters}{27}{subsection.2.2.3}
\contentsline {subsection}{\numberline {2.2.4}The $R^2$ Coefficient}{29}{subsection.2.2.4}
\contentsline {subsection}{\numberline {2.2.5}What to do if Errors \emph {Aren't} Normally Distributed?}{30}{subsection.2.2.5}
\contentsline {section}{\numberline {2.3}Feature Engineering}{32}{section.2.3}
\contentsline {subsection}{\numberline {2.3.1}Simple Feature Transformations}{33}{subsection.2.3.1}
\contentsline {subsection}{\numberline {2.3.2}Binary and Categorical Features: One-Hot Encodings}{33}{subsection.2.3.2}
\contentsline {subsubsection}{Categorical features}{34}{section*.23}
\contentsline {subsection}{\numberline {2.3.3}Missing Features}{37}{subsection.2.3.3}
\contentsline {subsection}{\numberline {2.3.4}Temporal Features}{38}{subsection.2.3.4}
\contentsline {subsection}{\numberline {2.3.5}Transformation of Output Variables}{39}{subsection.2.3.5}
\contentsline {section}{\numberline {2.4}Interpreting the Parameters of Linear Models}{40}{section.2.4}
\contentsline {section}{\numberline {2.5}Fitting Models with Gradient Descent}{41}{section.2.5}
\contentsline {subsection}{\numberline {2.5.1}Linear Regression via Gradient Descent}{43}{subsection.2.5.1}
\contentsline {section}{\numberline {2.6}Non-linear Regression}{43}{section.2.6}
\contentsline {subsection}{\numberline {2.6.1}Case Study: Image Popularity on Reddit}{44}{subsection.2.6.1}
\contentsline {section}{Exercises}{45}{section*.24}
\contentsline {section}{Project 1: Taxicab Tip Prediction (Part 1)}{46}{section*.25}
\contentsline {chapter}{\numberline {3}Classification and the Learning Pipeline}{48}{chapter.3}
\contentsline {section}{\numberline {3.1}Logistic Regression}{49}{section.3.1}
\contentsline {subsection}{\numberline {3.1.1}Fitting the Logistic Regressor}{50}{subsection.3.1.1}
\contentsline {subsection}{\numberline {3.1.2}Summary}{51}{subsection.3.1.2}
\contentsline {section}{\numberline {3.2}Other Classification Techniques}{51}{section.3.2}
\contentsline {section}{\numberline {3.3}Evaluating Classification Models}{53}{section.3.3}
\contentsline {subsection}{\numberline {3.3.1}Balanced Metrics for Classification}{54}{subsection.3.3.1}
\contentsline {subsection}{\numberline {3.3.2}Optimizing the Balanced Error Rate}{55}{subsection.3.3.2}
\contentsline {subsection}{\numberline {3.3.3}Using and Evaluating Classifiers for Ranking}{57}{subsection.3.3.3}
\contentsline {subsubsection}{Precision and recall}{57}{section*.26}
\contentsline {subsubsection}{$F_\beta $ score}{58}{section*.27}
\contentsline {subsubsection}{Precision and Recall @ $K$}{59}{section*.28}
\contentsline {subsubsection}{ROC and precision/recall curves}{59}{section*.29}
\contentsline {section}{\numberline {3.4}The Learning Pipeline}{61}{section.3.4}
\contentsline {subsection}{\numberline {3.4.1}Generalization, Overfitting and Underfitting}{61}{subsection.3.4.1}
\contentsline {subsubsection}{Overfitting}{62}{section*.30}
\contentsline {subsubsection}{Underfitting}{63}{section*.31}
\contentsline {subsection}{\numberline {3.4.2}Model Complexity and Regularization}{63}{subsection.3.4.2}
\contentsline {subsubsection}{Regularization}{64}{section*.32}
\contentsline {subsubsection}{Hyperparameters}{64}{section*.33}
\contentsline {subsubsection}{Fitting the regularized model (regression)}{65}{section*.34}
\contentsline {subsubsection}{Validation sets}{65}{section*.35}
\contentsline {subsubsection}{Why does the $\boldsymbol {\ell }_{\boldsymbol {1}}$ norm induce sparsity?}{67}{section*.36}
\contentsline {subsubsection}{`Theorems' regarding training, testing, and validation sets}{67}{section*.37}
\contentsline {subsection}{\numberline {3.4.3}Guidelines for Model Pipelines}{68}{subsection.3.4.3}
\contentsline {subsection}{\numberline {3.4.4}Regression and Classification in Tensorflow}{69}{subsection.3.4.4}
\contentsline {subsubsection}{Classification}{71}{section*.38}
\contentsline {section}{\numberline {3.5}Implementing the Learning Pipeline}{71}{section.3.5}
\contentsline {subsection}{\numberline {3.5.1}Significance Testing}{72}{subsection.3.5.1}
\contentsline {section}{Exercises}{75}{section*.39}
\contentsline {section}{Project 2: Taxicab Tip Prediction (Part 2)}{76}{section*.40}
\contentsline {part}{\MakeUppercase {Part\ TWO\relax \hskip 1em\relax Fundamentals of Personalized Machine Learning}}{77}{part.2}
\contentsline {chapter}{\numberline {4}Introduction to Recommender Systems}{79}{chapter.4}
\contentsline {section}{\numberline {4.1}Basic Setup and Problem Definition}{80}{section.4.1}
\contentsline {subsubsection}{How is recommendation different from regression or classification?}{81}{section*.41}
\contentsline {section}{\numberline {4.2}Representations for Interaction Data}{82}{section.4.2}
\contentsline {paragraph}{Activities as sets}{83}{section*.42}
\contentsline {paragraph}{Activities as matrices}{83}{section*.43}
\contentsline {section}{\numberline {4.3}Memory-based Approaches to Recommendation}{84}{section.4.3}
\contentsline {subsection}{\numberline {4.3.1}Defining a Similarity Function}{85}{subsection.4.3.1}
\contentsline {subsection}{\numberline {4.3.2}Jaccard Similarity}{86}{subsection.4.3.2}
\contentsline {subsection}{\numberline {4.3.3}Cosine Similarity}{88}{subsection.4.3.3}
\contentsline {paragraph}{Which similarity metric is `better'?}{90}{section*.44}
\contentsline {subsection}{\numberline {4.3.4}Pearson Similarity}{90}{subsection.4.3.4}
\contentsline {subsection}{\numberline {4.3.5}Using Similarity Measurements for Rating Prediction}{94}{subsection.4.3.5}
\contentsline {section}{\numberline {4.4}Random Walk Methods}{95}{section.4.4}
\contentsline {paragraph}{Relation to PageRank}{96}{section*.45}
\contentsline {section}{\numberline {4.5}Case Study: Amazon.com Recommendations}{98}{section.4.5}
\contentsline {section}{Exercises}{98}{section*.46}
\contentsline {section}{Project 3: A Recommender System for Books (Part 1)}{100}{section*.47}
\contentsline {chapter}{\numberline {5}Model-based Approaches to Recommendation}{102}{chapter.5}
\contentsline {subsubsection}{The Netflix Prize}{104}{section*.48}
\contentsline {section}{\numberline {5.1}Matrix Factorization}{104}{section.5.1}
\contentsline {subsubsection}{Relationship to the Singular Value Decomposition}{106}{section*.49}
\contentsline {subsection}{\numberline {5.1.1}Fitting the Latent Factor Model}{107}{subsection.5.1.1}
\contentsline {subsubsection}{User and item biases}{107}{section*.50}
\contentsline {subsubsection}{Gradient update equations}{109}{section*.51}
\contentsline {paragraph}{Other considerations for gradient descent}{109}{section*.52}
\contentsline {subsection}{\numberline {5.1.2}What Happened to User or Item Features?}{110}{subsection.5.1.2}
\contentsline {section}{\numberline {5.2}Implicit Feedback and Ranking Models}{110}{section.5.2}
\contentsline {subsection}{\numberline {5.2.1}Instance Re-weighting Schemes}{111}{subsection.5.2.1}
\contentsline {subsection}{\numberline {5.2.2}Bayesian Personalized Ranking}{112}{subsection.5.2.2}
\contentsline {section}{\numberline {5.3}`User-free' Model-based Approaches}{115}{section.5.3}
\contentsline {subsection}{\numberline {5.3.1}Sparse Linear Methods (SLIM)}{115}{subsection.5.3.1}
\contentsline {subsection}{\numberline {5.3.2}Factored Item Similarity Models (FISM)}{117}{subsection.5.3.2}
\contentsline {subsection}{\numberline {5.3.3}Item2Vec}{118}{subsection.5.3.3}
\contentsline {section}{\numberline {5.4}Evaluating Recommender Systems}{118}{section.5.4}
\contentsline {subsection}{\numberline {5.4.1}Precision and Recall @ $K$}{120}{subsection.5.4.1}
\contentsline {subsection}{\numberline {5.4.2}Mean Reciprocal Rank}{120}{subsection.5.4.2}
\contentsline {subsection}{\numberline {5.4.3}Cumulative Gain and NDCG}{121}{subsection.5.4.3}
\contentsline {subsection}{\numberline {5.4.4}Evaluation Metrics Beyond Model Accuracy}{122}{subsection.5.4.4}
\contentsline {section}{\numberline {5.5}Deep Learning for Recommendation}{122}{section.5.5}
\contentsline {subsection}{\numberline {5.5.1}Why the Inner Product?}{122}{subsection.5.5.1}
\contentsline {subsection}{\numberline {5.5.2}Multilayer Perceptron-based Recommendation}{124}{subsection.5.5.2}
\contentsline {subsubsection}{Neural Collaborative Filtering}{124}{section*.53}
\contentsline {subsection}{\numberline {5.5.3}Autoencoder-based Recommendation}{125}{subsection.5.5.3}
\contentsline {subsubsection}{AutoRec}{126}{section*.54}
\contentsline {subsection}{\numberline {5.5.4}Convolutional and Recurrent Networks}{127}{subsection.5.5.4}
\contentsline {subsection}{\numberline {5.5.5}How Effective are Deep Learning-Based Recommenders?}{128}{subsection.5.5.5}
\contentsline {section}{\numberline {5.6}Retrieval}{129}{section.5.6}
\contentsline {paragraph}{Euclidean distance}{129}{section*.55}
\contentsline {paragraph}{Inner product and cosine similarity}{129}{section*.56}
\contentsline {paragraph}{Approximate search and Jaccard similarity}{130}{section*.57}
\contentsline {section}{\numberline {5.7}Online Updates}{130}{section.5.7}
\contentsline {paragraph}{Regressing on $\boldsymbol {\gamma }_{\boldsymbol {u}}$ or $\boldsymbol {\gamma }_{\boldsymbol {i}}$}{130}{section*.58}
\contentsline {paragraph}{Cold-start and user-free models}{131}{section*.59}
\contentsline {paragraph}{Strategies for online training}{131}{section*.60}
\contentsline {section}{\numberline {5.8}Recommender Systems in Python with \emph {Surprise} and \emph {Implicit}}{131}{section.5.8}
\contentsline {subsection}{\numberline {5.8.1}Latent Factor Models (\emph {Surprise})}{131}{subsection.5.8.1}
\contentsline {subsection}{\numberline {5.8.2}Bayesian Personalized Ranking (\emph {Implicit})}{132}{subsection.5.8.2}
\contentsline {subsection}{\numberline {5.8.3}Implementing a Latent Factor Model in Tensorflow}{133}{subsection.5.8.3}
\contentsline {subsection}{\numberline {5.8.4}Bayesian Personalized Ranking in Tensorflow}{134}{subsection.5.8.4}
\contentsline {subsection}{\numberline {5.8.5}Efficient Batch-Based Optimization}{134}{subsection.5.8.5}
\contentsline {section}{\numberline {5.9}Beyond a `Black-Box' View of Recommendation}{135}{section.5.9}
\contentsline {section}{\numberline {5.10}History and Emerging Directions}{136}{section.5.10}
\contentsline {section}{Exercises}{138}{section*.61}
\contentsline {section}{Project 4: A Recommender System for Books (Part 2)}{139}{section*.62}
\contentsline {chapter}{\numberline {6}Content and Structure in Recommender Systems}{140}{chapter.6}
\contentsline {section}{\numberline {6.1}The Factorization Machine}{141}{section.6.1}
\contentsline {subsection}{\numberline {6.1.1}Neural Factorization Machines}{143}{subsection.6.1.1}
\contentsline {paragraph}{Wide and Deep learning for recommender systems}{143}{section*.63}
\contentsline {subsection}{\numberline {6.1.2}Factorization Machines in Python with \emph {FastFM}}{143}{subsection.6.1.2}
\contentsline {section}{\numberline {6.2}Cold-Start Recommendation}{145}{section.6.2}
\contentsline {subsection}{\numberline {6.2.1}Addressing Cold-Start Problems with Side Information}{146}{subsection.6.2.1}
\contentsline {subsection}{\numberline {6.2.2}Addressing Cold-Start Problems with Surveys}{147}{subsection.6.2.2}
\contentsline {section}{\numberline {6.3}Multisided Recommendation}{148}{section.6.3}
\contentsline {subsection}{\numberline {6.3.1}Online Dating}{148}{subsection.6.3.1}
\contentsline {subsection}{\numberline {6.3.2}Bartering Platforms}{149}{subsection.6.3.2}
\contentsline {section}{\numberline {6.4}Group- and Socially-Aware Recommendation}{151}{section.6.4}
\contentsline {subsection}{\numberline {6.4.1}Socially-Aware Recommendation}{151}{subsection.6.4.1}
\contentsline {subsection}{\numberline {6.4.2}Social Bayesian Personalized Ranking}{153}{subsection.6.4.2}
\contentsline {subsection}{\numberline {6.4.3}Group-Aware Recommendation}{155}{subsection.6.4.3}
\contentsline {subsection}{\numberline {6.4.4}Group Bayesian Personalized Ranking}{156}{subsection.6.4.4}
\contentsline {section}{\numberline {6.5}Price Dynamics in Recommender Systems}{157}{section.6.5}
\contentsline {subsection}{\numberline {6.5.1}Disentangling Prices and Preferences}{158}{subsection.6.5.1}
\contentsline {subsection}{\numberline {6.5.2}Estimating Willing-to-Pay Prices within Sessions}{159}{subsection.6.5.2}
\contentsline {subsection}{\numberline {6.5.3}Price Sensitivity and Price Elasticity}{160}{subsection.6.5.3}
\contentsline {section}{\numberline {6.6}Other Contextual Features in Recommendation}{163}{section.6.6}
\contentsline {subsection}{\numberline {6.6.1}Music and Audio}{163}{subsection.6.6.1}
\contentsline {subsection}{\numberline {6.6.2}Recommendation in Location-Based Networks}{164}{subsection.6.6.2}
\contentsline {subsection}{\numberline {6.6.3}Temporal, Textual, and Visual Features}{165}{subsection.6.6.3}
\contentsline {section}{\numberline {6.7}Online Advertising}{165}{section.6.7}
\contentsline {subsection}{\numberline {6.7.1}Matching Problems}{166}{subsection.6.7.1}
\contentsline {subsection}{\numberline {6.7.2}AdWords}{167}{subsection.6.7.2}
\contentsline {section}{Exercises}{168}{section*.64}
\contentsline {section}{Project 5: Cold-Start Recommendation on \emph {Amazon}}{170}{section*.65}
\contentsline {chapter}{\numberline {7}Temporal and Sequential Models}{172}{chapter.7}
\contentsline {section}{\numberline {7.1}Introduction to Regression with Time Series}{173}{section.7.1}
\contentsline {paragraph}{Autoregression}{173}{section*.66}
\contentsline {section}{\numberline {7.2}Temporal Dynamics in Recommender Systems}{175}{section.7.2}
\contentsline {subsection}{\numberline {7.2.1}Methods for Temporal Recommendation}{177}{subsection.7.2.1}
\contentsline {subsection}{\numberline {7.2.2}Case Study: Temporal Recommendation and the Netflix Prize}{178}{subsection.7.2.2}
\contentsline {subsection}{\numberline {7.2.3}What can Netflix Teach us about Temporal Models?}{182}{subsection.7.2.3}
\contentsline {section}{\numberline {7.3}Other Approaches to Temporal Dynamics}{183}{section.7.3}
\contentsline {subsection}{\numberline {7.3.1}Long-Term Dynamics of Opinions}{183}{subsection.7.3.1}
\contentsline {subsection}{\numberline {7.3.2}Short-Term Dynamics and Session-Based Recommendation}{184}{subsection.7.3.2}
\contentsline {subsection}{\numberline {7.3.3}User-Level Temporal Evolution}{185}{subsection.7.3.3}
\contentsline {section}{\numberline {7.4}Personalized Markov Chains}{186}{section.7.4}
\contentsline {paragraph}{Markov chains}{187}{section*.67}
\contentsline {section}{\numberline {7.5}Case Studies: Markov-Chain Models for Recommendation}{188}{section.7.5}
\contentsline {subsection}{\numberline {7.5.1}Factorized Personalized Markov Chains}{188}{subsection.7.5.1}
\contentsline {subsection}{\numberline {7.5.2}Socially-Aware Sequential Recommendation}{190}{subsection.7.5.2}
\contentsline {subsection}{\numberline {7.5.3}Locality-Based Sequential Recommendation}{191}{subsection.7.5.3}
\contentsline {subsection}{\numberline {7.5.4}Translation-Based Recommendation}{192}{subsection.7.5.4}
\contentsline {subsection}{\numberline {7.5.5}FPMC in \emph {Tensorflow}}{193}{subsection.7.5.5}
\contentsline {section}{\numberline {7.6}Recurrent Networks}{195}{section.7.6}
\contentsline {subsection}{\numberline {7.6.1}The Long Short-Term Memory Model}{197}{subsection.7.6.1}
\contentsline {section}{\numberline {7.7}Neural Network-Based Sequential Recommenders}{198}{section.7.7}
\contentsline {subsubsection}{Relationship to natural language processing}{198}{section*.68}
\contentsline {subsubsection}{`User-free' sequential recommendation}{199}{section*.69}
\contentsline {subsection}{\numberline {7.7.1}Recurrent Network-Based Recommendation}{200}{subsection.7.7.1}
\contentsline {subsection}{\numberline {7.7.2}Attention Mechanisms}{201}{subsection.7.7.2}
\contentsline {subsubsection}{Neural Attentive Recommendation}{201}{section*.70}
\contentsline {subsubsection}{Self-Attentive Sequential Recommendation}{201}{section*.71}
\contentsline {subsubsection}{BERT4Rec}{202}{section*.72}
\contentsline {subsubsection}{Attentional Factorization Machines}{203}{section*.73}
\contentsline {subsection}{\numberline {7.7.3}Summary}{203}{subsection.7.7.3}
\contentsline {section}{\numberline {7.8}Case Study: Personalized Heart-Rate Modeling}{204}{section.7.8}
\contentsline {section}{\numberline {7.9}History of Personalized Temporal Models}{205}{section.7.9}
\contentsline {section}{Exercises}{206}{section*.74}
\contentsline {section}{Project 6: Temporal and Sequential Dynamics in Business Reviews}{207}{section*.75}
\contentsline {part}{\MakeUppercase {Part\ THREE\relax \hskip 1em\relax Emerging Directions in Personalized Machine Learning}}{211}{part.3}
\contentsline {chapter}{\numberline {8}Personalized Models of Text}{213}{chapter.8}
\contentsline {section}{\numberline {8.1}Basics of Text Modeling: The Bag-of-Words Model}{214}{section.8.1}
\contentsline {subsection}{\numberline {8.1.1}Sentiment Analysis}{214}{subsection.8.1.1}
\contentsline {paragraph}{Removing capitalization and punctuation}{216}{section*.76}
\contentsline {paragraph}{Stemming}{216}{section*.77}
\contentsline {paragraph}{Stopwords}{216}{section*.78}
\contentsline {subsection}{\numberline {8.1.2}N-grams}{218}{subsection.8.1.2}
\contentsline {subsection}{\numberline {8.1.3}Word Relevance and Document Similarity}{221}{subsection.8.1.3}
\contentsline {subsection}{\numberline {8.1.4}Using TF-IDF for Search and Retrieval}{223}{subsection.8.1.4}
\contentsline {section}{\numberline {8.2}Distributed Word and Item Representations}{224}{section.8.2}
\contentsline {subsection}{\numberline {8.2.1}Item2Vec}{226}{subsection.8.2.1}
\contentsline {subsection}{\numberline {8.2.2}Word2Vec and Item2Vec with Gensim}{226}{subsection.8.2.2}
\contentsline {section}{\numberline {8.3}Personalized Sentiment and Recommendation}{228}{section.8.3}
\contentsline {subsection}{\numberline {8.3.1}Case Studies: Review-Aware Recommendation}{228}{subsection.8.3.1}
\contentsline {subsubsection}{Hidden Factors as Topics}{229}{section*.79}
\contentsline {subsubsection}{Other topic-modeling approaches}{230}{section*.80}
\contentsline {subsubsection}{Neural-network approaches}{231}{section*.81}
\contentsline {section}{\numberline {8.4}Personalized Text Generation}{231}{section.8.4}
\contentsline {subsubsection}{Why generate reviews?}{232}{section*.82}
\contentsline {subsection}{\numberline {8.4.1}RNN-based Review Generation}{233}{subsection.8.4.1}
\contentsline {subsubsection}{Conditional review generation}{233}{section*.83}
\contentsline {subsubsection}{Personalized review generation}{234}{section*.84}
\contentsline {subsection}{\numberline {8.4.2}Case Study: Personalized Recipe Generation}{236}{subsection.8.4.2}
\contentsline {subsection}{\numberline {8.4.3}Text-Based Explanations and Justifications}{237}{subsection.8.4.3}
\contentsline {paragraph}{Extractive vs.\nobreakspace {}abstractive approaches}{237}{section*.85}
\contentsline {subsubsection}{Crowd-sourced explanations}{237}{section*.86}
\contentsline {subsubsection}{Generating explanations from reviews}{238}{section*.87}
\contentsline {subsection}{\numberline {8.4.4}Conversational Recommendation}{238}{subsection.8.4.4}
\contentsline {subsubsection}{Query refinement}{239}{section*.88}
\contentsline {subsubsection}{Interactive recommendation}{239}{section*.89}
\contentsline {subsubsection}{Free-form conversation}{240}{section*.90}
\contentsline {section}{\numberline {8.5}Case Study: Google's \emph {Smart Reply}}{241}{section.8.5}
\contentsline {section}{Exercises}{242}{section*.91}
\contentsline {section}{Project 7: Personalized Document Retrieval}{243}{section*.92}
\contentsline {chapter}{\numberline {9}Personalized Models of Visual Data}{246}{chapter.9}
\contentsline {section}{\numberline {9.1}Personalized Image Search and Retrieval}{247}{section.9.1}
\contentsline {paragraph}{Latent factors}{247}{section*.93}
\contentsline {paragraph}{Joint embeddings}{247}{section*.94}
\contentsline {section}{\numberline {9.2}Visually-Aware Recommendation and Personalized Ranking}{248}{section.9.2}
\contentsline {subsection}{\numberline {9.2.1}Visual Bayesian Personalized Ranking}{248}{subsection.9.2.1}
\contentsline {subsubsection}{Modeling the visual evolution of fashion trends}{250}{section*.95}
\contentsline {section}{\numberline {9.3}Case Studies: Visual and Fashion Compatibility}{251}{section.9.3}
\contentsline {subsection}{\numberline {9.3.1}Estimating Compatibility from Co-purchases}{252}{subsection.9.3.1}
\contentsline {subsection}{\numberline {9.3.2}Learning Compatibility from Images in the Wild}{254}{subsection.9.3.2}
\contentsline {subsection}{\numberline {9.3.3}Generating Fashionable Wardrobes}{255}{subsection.9.3.3}
\contentsline {subsection}{\numberline {9.3.4}Domains other than Fashion}{256}{subsection.9.3.4}
\contentsline {subsection}{\numberline {9.3.5}Other Techniques for Substitutable and Complementary Product Recommendation}{257}{subsection.9.3.5}
\contentsline {subsubsection}{Learning non-metric item relationships}{257}{section*.96}
\contentsline {subsubsection}{Diversifying complementary item recommendation}{258}{section*.97}
\contentsline {subsubsection}{Incorporating item types}{259}{section*.98}
\contentsline {subsection}{\numberline {9.3.6}Implementing a Compatibility Model in \emph {Tensorflow}}{259}{subsection.9.3.6}
\contentsline {section}{\numberline {9.4}Personalized Generative Models of Images}{261}{section.9.4}
\contentsline {section}{Exercises}{263}{section*.99}
\contentsline {section}{Project 8: Generating Compatible Outfits}{264}{section*.100}
\contentsline {chapter}{\numberline {10}The Consequences of Personalized Machine Learning}{267}{chapter.10}
\contentsline {section}{\numberline {10.1}Measuring Diversity}{269}{section.10.1}
\contentsline {section}{\numberline {10.2}Filter Bubbles, Diversity, and Extremification}{271}{section.10.2}
\contentsline {subsection}{\numberline {10.2.1}Exploring Diversity Through Simulation}{271}{subsection.10.2.1}
\contentsline {subsection}{\numberline {10.2.2}Empirically Measuring Recommendation Diversity}{271}{subsection.10.2.2}
\contentsline {subsection}{\numberline {10.2.3}Auditing Pathways to Extreme Content}{272}{subsection.10.2.3}
\contentsline {section}{\numberline {10.3}Diversification Techniques}{273}{section.10.3}
\contentsline {subsection}{\numberline {10.3.1}Maximal Marginal Relevance}{273}{subsection.10.3.1}
\contentsline {subsection}{\numberline {10.3.2}Other Re-ranking Approaches to Diverse Recommendation}{274}{subsection.10.3.2}
\contentsline {subsection}{\numberline {10.3.3}Determinantal Point Processes}{276}{subsection.10.3.3}
\contentsline {section}{\numberline {10.4}Implementing a Diverse Recommender}{277}{section.10.4}
\contentsline {paragraph}{Examples of diversified recommendations}{279}{section*.101}
\contentsline {section}{\numberline {10.5}Case Studies on Recommendation and Consumption Diversity}{280}{section.10.5}
\contentsline {subsection}{\numberline {10.5.1}Diversity on Spotify}{280}{subsection.10.5.1}
\contentsline {subsubsection}{Guiding users to more diverse content}{281}{section*.102}
\contentsline {subsection}{\numberline {10.5.2}Filter Bubbles and Online News Consumption}{282}{subsection.10.5.2}
\contentsline {subsubsection}{Diversity across consumption channels}{282}{section*.103}
\contentsline {subsubsection}{Filter bubbles on Google News}{283}{section*.104}
\contentsline {section}{\numberline {10.6}Other Metrics Beyond Accuracy}{284}{section.10.6}
\contentsline {subsection}{\numberline {10.6.1}Serendipity}{285}{subsection.10.6.1}
\contentsline {subsubsection}{Serendipity in music recommendation}{285}{section*.105}
\contentsline {subsubsection}{Investigating serendipity via user studies}{286}{section*.106}
\contentsline {subsection}{\numberline {10.6.2}Unexpectedness}{287}{subsection.10.6.2}
\contentsline {subsection}{\numberline {10.6.3}Calibration}{288}{subsection.10.6.3}
\contentsline {section}{\numberline {10.7}Fairness}{289}{section.10.7}
\contentsline {subsection}{\numberline {10.7.1}Multisided Fairness}{291}{subsection.10.7.1}
\contentsline {subsection}{\numberline {10.7.2}Implementing Fairness Objectives in \emph {Tensorflow}}{293}{subsection.10.7.2}
\contentsline {section}{\numberline {10.8}Case Studies on Gender Bias in Recommendation}{294}{section.10.8}
\contentsline {subsection}{\numberline {10.8.1}Data Resampling and Popularity Bias}{294}{subsection.10.8.1}
\contentsline {subsection}{\numberline {10.8.2}Bias and Author Gender in Book Recommendations}{295}{subsection.10.8.2}
\contentsline {subsection}{\numberline {10.8.3}Gender Bias in Marketing}{295}{subsection.10.8.3}
\contentsline {section}{Exercises}{297}{section*.107}
\contentsline {section}{Project 9: Diverse and Fair Recommendations}{298}{section*.108}