#026 ML Виталий Моисеев. Рекомендательные системы - как вырастить "счастье пользователей"

Item: #026 ML Виталий Моисеев. Рекомендательные системы - как вырастить "счастье пользователей"
Rating: 4
Author: PostSingularity Institute

#026 ML Vitaly Moiseev. Recommender systems - how to grow "user happiness"

by Machine Learning Podcast · View the channel on PSI

Formal & Physical Sciences AI — Models & Research UYQArtificial intelligence (AI)UYQMMachine learning

🎙 Machine Learning Podcast 👥 5K 📅 June 13, 2021 ⏱ 65 min 👁 810 📄 expert opinion 🧭 2026-10-09

Available in: English (current) Français

Keywords

recommendation systemscollaborative filteringembeddingscold startexploration vs exploitation

Summary

The podcast episode features Vitaly Moiseev, head of video recommendation quality at Yandex, discussing recommendation systems. It begins with his background and role as an analyst, emphasizing the importance of metrics as approximations of user happiness. The core techniques are explained: from heuristic rules to collaborative filtering and matrix factorization, producing user and item embeddings. The cold start problem for new items and new users is a major challenge; neural networks help generate embeddings from content and user history. The episode also covers exploration vs exploitation, the need to balance familiar and novel recommendations to avoid user churn. Yandex’s Toloka platform for crowdsourced data labeling is mentioned as a tool for offline evaluation. The latter part discusses internship opportunities at Yandex and the skills needed. Throughout, the conversation provides an accessible yet detailed overview for non-specialists, enriched by the guest’s practical experience.

142 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video offers substantial value by demystifying recommendation systems through concrete examples and analogies. The guest explains complex ideas like collaborative filtering and embeddings with clarity, grounding them in real-world applications (e.g., Yandex, TikTok). The argumentation is coherent and evidence-based, drawing from industrial practice. The discussion of cold start and exploration-exploitation trade-offs highlights nuanced challenges beyond simple algorithmic accuracy. The guest’s dual role as an analyst and manager lends credibility to the practical insights on metrics and user satisfaction.

Scientific Rigor, Source Quality, Title Accuracy

The rigour is moderate: the conversation is informal but based on well-established concepts in the field. No on-screen citations are given, but the description provides relevant links (e.g., Recommender Systems Handbook, a Habr article on the Netflix Prize, Yandex Toloka). The title accurately reflects the content, focusing on how recommendation systems aim to ‘grow user happiness’. The episode fulfills its promise, delivering an educational dialogue that remains technically credible without being over-specialized.

166 words

Title / Content Match

The title accurately reflects the content: the episode dives into how recommendation systems work and their effect on user satisfaction, with a focus on growing 'happiness'.

Quality & Reliability

7/10

The podcast features an expert from Yandex with deep practical knowledge of recommendation systems. The discussion covers core concepts clearly, but lacks formal citations on-screen; reliance on the guest's expertise and the provided description links helps anchor the content. A solid educational resource, though not peer-reviewed.

Key Moments

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

Cited Sources

Recommender Systems Handbook — Mentioned as a comprehensive book on recommendation techniques.
Феерический финал конкурса Netflix (Habr article) — Referenced in the description as a narrative about the Netflix Prize competition.
Yandex Toloka — Crowdsourcing platform used for data labeling and quality evaluation.
Дзен-митап: исследования и рекомендательные системы (YouTube video) — Related meetup on recommendation systems mentioned by the guest.
Стажировки в Яндексе — Information about Yandex internships, relevant to the final part of the episode.
Machine Learning Podcast episode 026 on Anchor — Official podcast episode page.

Concurring Sources

Recommender Systems Handbook — Supports the technical concepts of collaborative filtering and embeddings discussed.
Дзен-митап: исследования и рекомендательные системы — Related discussion on recommendation research that aligns with the episode's themes.

External References

Contribution & Novelties

The podcast provides an accessible, practitioner-oriented explanation of recommendation systems, focusing on industrial challenges like cold start and exploration-exploitation trade-offs. It bridges theory and practice by drawing on Yandex’s real-world deployments, offering insights not typically found in academic surveys.

Pour aller plus loin :

Collaborative filtering (Wikipedia) — Foundational concept discussed at length.
Matrix factorization (recommender systems) (Wikipedia) — Core mathematics behind embeddings.
Cold start (recommender systems) (Wikipedia) — Central challenge emphasized in the episode.
Exploration-exploitation dilemma (Wikipedia) — Key trade-off in recommendation quality.

84 words

Radar Profile

The profile shows a balanced and solid educational resource: quantity and quality of information are high (8/10 each), while technical depth is moderate (6/10), making it accessible to a broad audience. Reliability is strong (7/10) due to the expert guest and supporting links, though it lacks formal scholarly rigour.

Reliability 7/10