[ИАД, весна 2026] Рекомендательные системы. Лекция 2

Value of the Information & Strength of the Argument

The lecture provides a solid foundation in collaborative filtering, clearly explaining the core concepts and mathematical formulations. The instructor effectively contrasts memory-based and model-based approaches, and the progression from simple methods like Item KNN to more sophisticated ones like SLIM and EASE is logical. The argumentation is coherent, with each model building on the previous, and the practical session reinforces the theoretical concepts. However, the lecture lacks critical evaluation of the methods’ strengths and weaknesses, and does not discuss potential pitfalls or alternative approaches in depth.

Scientific Rigor, Source Quality, Title Accuracy

The lecture demonstrates scientific rigor by referencing classic literature, such as the book by Ricci et al., and seminal papers on SLIM and EASE. The instructor also mentions historical systems like Tapestry, providing context. The title accurately reflects the content, as it is the second lecture in a course on recommender systems, focusing on collaborative filtering. The sources cited are appropriate and credible, though the lecture does not critically assess them. The practical session uses the well-known MovieLens dataset, which is a standard benchmark in the field.

The title accurately reflects the content, as it is the second lecture in a course on recommender systems, focusing on collaborative filtering.

8/10

The lecture provides a structured, academic overview of collaborative filtering methods, with clear definitions, mathematical formulations, and references to classic literature. The content is consistent with established knowledge in the field, though it lacks empirical validation or critical discussion of limitations.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction to the lecture and definition of collaborative filtering.
• 05:00Historical context: Tapestry system and early definitions.
• 10:00Memory-based vs. model-based approaches.
• 15:00User-based vs. item-based methods and their characteristics.
• 20:00Formulation of rating prediction using similarity-weighted averages.
• 25:00Similarity measures: cosine, Pearson correlation, and practical adjustments like IDF and shrinkage.
• 30:00Item-based KNN model and its formula.
• 35:00Introduction to SLIM and EASE models.
• 40:00Optimization problem for SLIM and EASE, including constraints and regularization.
• 45:00Derivation of closed-form solution for EASE using Lagrange multipliers.
• 50:00Practical session: MovieLens dataset, data splitting, and evaluation metrics.

• Recommender Systems Handbook — Referenced as a classic book on recommender systems.
• SLIM: Sparse Linear Methods for Top-N Recommender Systems — Referenced as the source for the SLIM model.
• EASE: Embarrassingly Shallow Autoencoders for Sparse Data — Referenced as the source for the EASE model.

• Recommender Systems Handbook — The lecture's content aligns with standard knowledge in the field.