Recommender Systems and Graph Theory Concepts

Vocabulary flashcards covering key concepts in recommender systems, matrix factorization, feedback types, distance metrics, and community detection in graphs.

User–Item Rating Matrix

A matrix where each row is a user, each column is an item, and entries contain ratings when available.

Missing Values

Unobserved user–item interactions represented as empty cells in the rating matrix.

Sparsity

The condition where most user–item ratings are missing, leading to a sparse matrix.

Impact of Sparsity

Makes similarity estimates noisy because users share few co-rated items.

Long-Tail Distribution

Popular items form a small 'head,' while many niche items form a large 'tail.'

User–User Collaborative Filtering

Predict ratings using users who have similar rating patterns.

Item–Item Collaborative Filtering

Predict ratings using items that are similar to those a user has rated.

Neighborhood

A set of users or items deemed similar based on similarity metrics.

Cold-Start Problem

Difficulty recommending items or users with insufficient historical data.

Latent Factors

Low-dimensional vectors representing hidden traits of users and items.

Global Mean

Overall average rating across all user–item pairs.

User Bias

Tendency of a user to rate higher or lower than average.

Item Bias

Tendency of an item to receive higher or lower ratings than average.

Latent Space

The learned embedding space where users and items are represented as vectors.

RMSE

Rating prediction error measuring average squared deviation between predicted and true ratings.

Precision@k

Fraction of top-k recommended items that are relevant.

Recall@k

Fraction of relevant items that appear in the top-k recommendations.

Hit-Rate

Whether at least one relevant item appears in the recommendations.

NDCG@k

Ranking metric that assigns higher weight to correctly ranked relevant items near the top.

Rating Prediction Task

Predicting explicit numerical ratings.

Ranking Task

Ordering items by predicted relevance rather than predicting exact rating values.

Explicit Feedback

Direct user-provided ratings or evaluations.

Implicit Feedback

Behavioral signals such as clicks, views, or watch time.

Noisy Feedback

Implicit signals that do not directly reflect true preference strength.

Exposure Bias

Observed behavior depends on what users were shown, not all available items.

Position Bias

Higher-ranked items receive more attention regardless of true relevance.

Missing-Not-Negative

Absence of interaction is not the same as disliking an item.

Similarity Measure

Quantifies how similar two users or items are (e.g., cosine similarity).

Cosine Similarity

Measures angle-based similarity between vectors.

Euclidean Distance

Measures straight-line distance between vectors.

Curse of Dimensionality

Distance metrics become less meaningful in high-dimensional spaces.

Feature Scaling

Adjusting feature magnitudes to ensure equal influence in distance-based models.

k Value (k-NN)

Number of neighbors; low k risks high variance, high k risks high bias.

Linear Separability

Existence of a linear boundary that perfectly separates classes.

Perceptron Convergence

The perceptron converges only if the data is linearly separable.

Decision Boundary

A hyperplane that divides classes.

Order Dependence

Perceptron updates depend on the sequence of training examples.

Margin

Distance between the decision boundary and the nearest data points.

Support Vectors

Data points that lie on or near the margin and define the decision boundary.

Soft-Margin SVM

Allows some misclassification to improve generalization.

Kernel Trick

Method for learning non-linear boundaries by computing similarity in transformed feature spaces.

RBF Kernel

A popular kernel that measures similarity based on distance in feature space.

Overfitting

Model fits training data too closely and performs poorly on unseen data.

Underfitting

Model is too simple and fails to capture important patterns.

Train/Validation/Test Split

Partitioning data to train, tune, and evaluate a model.

Cross-Validation

Repeated training/testing on multiple splits for more reliable evaluation.

Regularization

Penalizing model complexity to reduce overfitting.

L2 Regularization

Penalizes large parameter values via squared magnitude.

Community

Group of nodes densely connected internally and sparsely connected externally.

Modularity

Metric evaluating how well a division separates dense communities.

Modularity Resolution Limit

Modularity may fail to detect small but real communities.

Edge Betweenness

Number of shortest paths that pass through an edge.

Girvan–Newman Algorithm

Detects communities by iteratively removing edges with highest betweenness.

Bridge Edge

An edge whose removal disconnects parts of the network.

Bottleneck

Node or edge that many shortest paths depend on.

Affiliation Graph Model

Model where nodes belong to multiple communities and connect based on shared memberships.

Overlapping Communities

Communities where nodes can have more than one membership.

Overlap Region

Area where nodes with multiple affiliations show higher connectivity.

Connection Probability

Likelihood that two nodes connect increases with the number of shared affiliations.