Blog Post
Recommendation Systems: Papers to Read
A bottom-up reading list for understanding recommendation systems: from classic CF through industrial-scale deep learning, sequential models, graph networks, and LLM-era generative rec. I'll keep adding to this over time rather than starting a new post for every batch.
Type key: paper = peer-reviewed / arxiv preprint · blog = blog/post · report = technical report · book = textbook/chapter · survey = survey paper · website = tool/org site · dataset = data release
Part 1 - Foundations & Classic Methods
Collaborative Filtering Origins
- GroupLens: Applying Collaborative Filtering to Usenet News - Resnick et al. (1994) · first CF paper ·
paper - Amazon.com Recommendations: Item-to-Item Collaborative Filtering - Linden, Smith & York, Amazon (2003) · powers "customers who bought" ·
paper - Collaborative Filtering for Implicit Feedback Datasets - Hu, Koren & Volinsky (2008) · WMF: treating clicks as confidence, not preference ·
paper - Matrix Factorization Techniques for Recommender Systems - Koren, Bell & Volinsky (2009) · the canonical MF overview from Netflix Prize ·
paper - SVD++: Factorization Meets the Neighborhood - Koren (2008) · integrates implicit + explicit feedback ·
paper - Lessons from the Netflix Prize - BellKor (2009) · ensemble + SVD++ winning writeup ·
report - BPR: Bayesian Personalized Ranking from Implicit Feedback - Rendle et al. (2009) · pairwise ranking loss for implicit data ·
paper
Factorization Machines & Feature Interaction
- Factorization Machines - Rendle (2010) · unifies MF + polynomial feature interaction ·
paper - Field-aware Factorization Machines (FFM) - Juan et al. (2016) · per-field interaction parameters ·
paper
Simple Baselines Worth Knowing
- SLIM: Sparse Linear Methods for Top-N Recommendation - Ning & Karypis (2011) · learned item-item sparse weight matrix ·
paper - EASE^R: Embarrassingly Shallow Autoencoders for Sparse Data - Steck (2019) · closed-form linear model that beats most deep models ·
paper - NMF: Algorithms for Non-negative Matrix Factorization - Lee & Seung, Nature (1999) · parts-based decomposition; used in topic rec ·
paper
Part 2 - Deep Learning for Ranking & CTR Prediction
- Wide & Deep Learning for Recommender Systems - Cheng et al., Google (2016) · memorization (wide) + generalization (deep) ·
paper - DeepFM: A Factorization-Machine based Neural Network for CTR - Guo et al., Huawei (2017) · FM + MLP, no manual feature engineering ·
paper - Neural Collaborative Filtering (NCF) - He et al. (2017) · MLP replaces dot product in MF ·
paper - Deep & Cross Network (DCN) - Wang et al., Google (2017) · bounded-degree feature crosses explicitly ·
paper - DCN V2: Improved Deep & Cross Network - Wang et al., Google (2021) · matrix cross layers; state-of-the-art on Criteo ·
paper - xDeepFM: Combining Explicit and Implicit Feature Interactions - Lian et al., Microsoft (2018) · compressed interaction network ·
paper - DIN: Deep Interest Network for CTR Prediction - Zhou et al., Alibaba (2018) · attention over user history, target-aware ·
paper - DIEN: Deep Interest Evolution Network - Zhou et al., Alibaba (2019) · GRU with attention-based gate for evolving interest ·
paper - AutoInt: Automatic Feature Interaction Learning via Self-Attentive Neural Networks - Song et al. (2019) · multi-head self-attention on feature embeddings ·
paper - FiBiNET: Feature Importance and Bilinear Feature Interaction - Huang et al., Sina Weibo (2019) · SENET for feature importance + bilinear ·
paper - BST: Behavior Sequence Transformer for E-commerce Recommendation - Chen et al., Alibaba (2019) · transformer applied to user behavior for CTR ·
paper - DLRM: Deep Learning Recommendation Model - Naumov et al., Meta (2019) · open-source reference industrial RecSys architecture ·
paper - FinalMLP: An Enhanced Two-Stream MLP Model for CTR Prediction (2023) · simple dual-stream MLP beats complex models ·
paper - DHEN: Deep Hierarchical Ensemble Network for RecSys - Meta (2022) · hierarchical ensemble of interaction modules ·
paper - CTRL: Connect Tabular and Language Model for CTR Prediction - Yuan et al., Alibaba (2023) · bridges structured feature tables and LLM text for industrial CTR ·
paper
Autoencoders for CF
- Collaborative Denoising Auto-Encoders (CDAE) - Wu et al. (2016) ·
paper - Variational Autoencoders for Collaborative Filtering (Mult-VAE) - Liang et al. (2018) · principled generative model for implicit CF ·
paper
Part 3 - Sequential & Session-Based Recommendation
- Session-based Recommendations with Recurrent Neural Networks (GRU4Rec) - Hidasi et al. (2015) · first RNN-based session recommendations ·
paper - Improved GRU4Rec with Pairwise Loss - Hidasi & Karatzoglou (2018) ·
paper - Caser: Personalized Top-N Sequential Recommendation via Convolutional Sequence Embedding - Tang & Wang (2018) · CNN for short-term sequential patterns ·
paper - SASRec: Self-Attentive Sequential Recommendation - Kang & McAuley (2018) · causal transformer over item history; dominant baseline ·
paper - BERT4Rec: Sequential Recommendation with Bidirectional Encoder Representations - Sun et al. (2019) · cloze-task masked training for sequential rec ·
paper - SSE-PT: Sequential Recommendation via Personalized Transformer (2020) · personalization tokens in attention ·
paper - FMLP-Rec: Filter-enhanced MLP for Sequential Recommendation with Noise Filtering (2022) · learnable filters beat attention in many settings ·
paper - FEARec: Frequency Enhanced Hybrid Attention for Sequential Recommendation (2023) · frequency domain + time domain attention ·
paper - E4SRec: Effective, Efficient, Extensible Sequential Rec with LLMs (2024) · practical LLM-for-SeqRec deployable in production; LoRA + hybrid ID+text ·
paper - Breaking the Length Barrier: LLMs Excel at Supervised Long-Context Recommendation (2024) · fine-tuned LLMs with 10k+ behavior histories outperform all ID baselines ·
paper
Part 4 - Graph-Based Recommendation
- Graph Convolutional Matrix Completion (GC-MC) - Kipf et al. (2017) · GCN on bipartite user-item graph for MF ·
paper - PinSage: Graph Convolutional Neural Networks for Web-Scale Recommender Systems - Ying et al., Pinterest (2018) · random walk + GCN at 3B item production scale ·
paper - NGCF: Neural Graph Collaborative Filtering - Wang et al. (2019) · propagates user-item interactions via high-order graph ·
paper - LightGCN: Simplifying and Powering Graph Convolution for Recommendation - He et al. (2020) · removes weight matrices + nonlinearity; dominant baseline ·
paper - UltraGCN: Ultra Simplification of Graph Convolutional Networks (2021) · constraint-based loss, skips aggregation; faster than LightGCN ·
paper - SGL: Self-supervised Graph Learning for Recommendation - Wu et al. (2021) · three graph augmentation types + contrastive loss ·
paper - SimGCL: Are Graph Augmentations Necessary? (2022) · uniform noise in embedding space beats graph augmentation ·
paper - NCL: Neighborhood-enriched Contrastive Learning for Collaborative Filtering (2022) · structural and semantic neighbors as contrastive pairs ·
paper
Part 5 - Two-Tower Retrieval Models
- Deep Neural Networks for YouTube Recommendations - Covington, Adams & Basilic, Google (2016) · two-stage retrieval + ranking; the industry blueprint ·
paper - Learning Deep Structured Semantic Models (DSSM) - Huang et al., Microsoft (2013) · dual encoder text matching; predecessor to two-tower ·
paper - Sampling-Bias-Corrected Neural Modeling for Large Corpus Item Recommendations - Yi et al., Google (2019) · in-batch negatives with frequency correction for retrieval ·
paper - Embedding-based Retrieval in Facebook Search - Huang et al., Meta (2020) · hard negative mining + quantization at trillion scale ·
paper - Mixed Negative Sampling for Learning Two-Tower Neural Networks - Yi et al., Google (2019) · batch + random negatives combined ·
paper - MOBIUS: Towards the Next Generation of Query-Ad Matching in Baidu - Fan et al., Baidu (2019) · active learning for retrieval ·
paper
ANN / Retrieval Infrastructure
- Product Quantization for Nearest Neighbor Search - Jégou et al. (2011) · core ANN compression technique ·
paper - FAISS: A Library for Efficient Similarity Search - Douze et al., Meta (2024) · reference library for billion-scale MIPS ·
paper - HNSW: Efficient and Robust Approximate Nearest Neighbor Search - Malkov & Yashunin (2016) · graph-based ANN index; production standard ·
paper - ScaNN: Accelerating Large-Scale Inference with Anisotropic Vector Quantization - Guo et al., Google (2020) · MIPS-optimized quantization, beats FAISS ·
paper
Part 6 - Multi-Task Learning & Multi-Objective Optimization
- MMoE: Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts - Ma et al., Google (2018) · gated expert sharing for multi-task rec ·
paper - ESMM: Entire Space Multi-task Model for CVR Prediction - Ma et al., Alibaba (2018) · CVR via CTR×CTCVR, corrects sample selection bias ·
paper - Recommending What Video to Watch Next: A Multitask Ranking System - Zhao et al., Google (2019) · MMoE for Watch Next + position debiasing via shallow tower ·
paper - PLE: Progressive Layered Extraction for Multi-Task Learning - Tang et al., Tencent (2020) · separates shared and task-specific experts explicitly ·
paper - SNR: Sub-Network Routing for Flexible Parameter Sharing in Multi-task Learning - Ma et al. (2019) ·
paper
Part 7 - Long User History & Memory
- MIMN: Practice on Long Sequential User Behavior Modeling for CTR Prediction - Pi et al., Alibaba (2019) · memory network for 1000+ item history ·
paper - SIM: Search-based User Interest Model - Pi et al., Alibaba (2020) · two-stage hard/soft search for 10k+ length histories ·
paper - ETA: End-to-end Target Attention for Long Sequence Modeling - Chen et al. (2021) · hash-based target attention over long history ·
paper - SDIM: Sampling-based Deep Interaction Model for Long Sequence CTR - Cao et al., Meituan (2022) ·
paper
Part 8 - Industrial Systems at Scale
- Deep Neural Networks for YouTube Recommendations - Covington et al., Google (2016) ·
paper - DLRM: Deep Learning Recommendation Model - Naumov et al., Meta (2019) ·
paper - Pixie: A System for Recommending 3+ Billion Items to 200M+ Users in Real-Time - Eksombatchai et al., Pinterest (2018) · bipartite random walk at scale ·
paper - Real-time Personalization using Embeddings for Search Ranking at Airbnb - Grbovic & Cheng, Airbnb (2018) · listing2vec via listing/user co-click sessions ·
paper - Applying Deep Learning To Airbnb Search - Haldar et al., Airbnb (2019) · neural ranking with online/offline gap analysis ·
paper - Improving Deep Learning for Airbnb Search - Haldar et al., Airbnb (2020) · Listing Quality Score + architecture lessons ·
paper - Monolith: Real-Time Recommendation System with Collisionless Embedding Table - Liu et al., ByteDance/TikTok (2022) · streaming training on live traffic, no collision hashing ·
paper - Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Generative Recommendations (HSTU) - Zhai et al., Meta (2024) · 1T parameter HSTU architecture; scaling laws for rec ·
paper - Wukong: Towards a Scaling Law for Large-Scale Recommendation - Yao et al., LinkedIn (2024) · how rec models scale vs LLMs ·
paper - PinnerFormer: Sequence Modelling for User Representation at Pinterest - Zhai et al., Pinterest (2022) · offline-trained transformer for dense user embeddings ·
paper - On the Factory Floor: ML Engineering for Industrial-Scale Ads Recommendation Models - Meta (2022) · engineering lessons at ads-scale ·
paper - Instagram's Explore Recommender System - Meta AI (2019) ·
blog - Twitter/X Open-Sourced Recommendation Algorithm - Twitter/X (2023) ·
blog - Netflix Recommendation: Beyond the 5 Stars - Netflix Tech Blog (2012) ·
blog - YouTube Blog: How the Recommendation System Works - YouTube (2021) ·
blog - Eugene Yan's Applied RecSys Writing - practitioner deep-dives on production rec ·
blog
Part 9 - Knowledge Graph-Enhanced Recommendation
- RippleNet: Propagating User Preferences on the Knowledge Graph - Wang et al. (2018) · ripple set propagation over KG triples ·
paper - KGCN: Knowledge Graph Convolutional Networks for Recommendation - Wang et al. (2019) ·
paper - KGAT: Knowledge Graph Attention Network for Recommendation - Wang et al. (2019) · attentive graph propagation in KG ·
paper - MKR: Multi-Task Feature Learning for Knowledge Graph Enhanced Recommendation - Wang et al. (2019) · cross-compress units bridge KG and rec ·
paper - KGIN: Knowledge Graph-based Intent Network for Recommendation (2021) · user intent modeling via KG relational paths ·
paper
Part 10 - Contrastive & Self-Supervised Learning for RecSys
- CL4SRec: Contrastive Learning for Sequential Recommendation - Xie et al. (2022) · crop/mask/reorder augmentations on sequences ·
paper - CoSeRec: Contrastive Self-supervised Sequential Recommendation with Robust Augmentation (2021) ·
paper - SGL: Self-supervised Graph Learning for Recommendation - Wu et al. (2021) · graph augmentation + contrastive loss ·
paper - DirectAU: Towards Representation Alignment and Uniformity in Collaborative Filtering (2022) · alignment + uniformity objectives from Wang & Isola ·
paper - Alignment and Uniformity on the Hypersphere - Wang & Isola (2020) · theoretical grounding for CL objectives ·
paper - NCL: Neighborhood-enriched Contrastive Learning (2022) · structural and semantic neighborhood as positive pairs ·
paper
Part 11 - Multimodal Recommendation
- VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback - He & McAuley (2016) · visual features from CNN in MF ·
paper - ACF: Attentive Collaborative Filtering with Item and Component-level Attention - Chen et al. (2017) ·
paper - BM3: Bootstrapped Multi-modal Self-supervised Learning for Recommendation (2022) · modal dropout + contrastive learning ·
paper - MMGCN: Multi-modal Graph Convolution Network for Personalized Recommendation (2021) ·
paper - Artwork Personalization at Netflix - Netflix Tech Blog (2017) · personalized thumbnail selection via contextual bandits ·
blog - AlphaRec: Scalable Multimodal Recommendation with Vision-Language Models (2025) · CLIP embeddings as item representations; zero-shot cold-start rec ·
paper
Part 12 - LLMs & Generative Recommendation
- P5: Pretrain, Personalize, Prompt - Towards a Unified Paradigm for RecSys - Geng et al. (2022) · text-to-text unified LLM for all RecSys tasks ·
paper - TIGER: Generative Retrieval via Semantic IDs - Rajput et al., Google (2023) · autoregressive generation of hierarchical item IDs ·
paper - How to Index Item IDs for Recommendation Foundation Models (2023) · ID vs text indexing comparison ·
paper - TALLRec: Align LLM with Recommendation via Instruction Tuning (2023) · LoRA two-stage tuning for LLM rec ·
paper - InstructRec: Recommendation as Instruction Following (2023) · instruction-tuning for personalized rec ·
paper - LLaRA: Large Language-Recommendation Assistant (2023) · hybrid prompts with collaborative + text features ·
paper - RLMRec: Representation Learning with LLMs for Recommendation (2023) · LLM-generated semantic profiles augment CF ·
paper - Is ChatGPT a Good Recommender? A Preliminary Study (2023) · zero-shot LLM rec evaluation ·
paper - BIGRec: Grounding Language Models for Recommendation (2023) · grounding LLM outputs to real catalog items ·
paper - Actions Speak Louder than Words (HSTU) - Zhai et al., Meta (2024) · generative rec at trillion-scale ·
paper - A Survey on Large Language Models for Recommendation - Wu et al. (2023) · covers prompting, tuning, and agent-based rec ·
survey - RecSys in the Era of Large Language Models: A Survey (2023) ·
survey - NoteLLM: A Retrievable Large Language Model for Note Recommendation - Xiaohongshu/REDNote, WWW (2024) · end-to-end LLM as retriever at production note-rec scale ·
paper - HLLM: Enhancing Sequential Recommendations via Hierarchical Large Language Models (2024) · item-level LLM encodes content; user-level LLM models behavioral history ·
paper - AgentCF: Collaborative Learning with Autonomous Language Agents (2024) · LLM agents simulate user-item interaction to generate training signal ·
paper - RecMind: Large Language Model Powered Agent for Recommendation (2024) · planning + memory + tool use in LLM agent for rec ·
paper - AgentRec: Simulation-based Evaluation via LLM Agents (2024) · replaces offline user datasets with LLM-simulated users for eval ·
paper - MoRec: Recommenders are Sequentially Human Preference Models (Text vs ID) (2022, widely discussed 2024) · text encoders rival ID embeddings - live frontier debate ·
paper - Generative Recommendation: Towards Next-generation Recommender Paradigm (2023) · position paper framing the shift from retrieve-rank to generate ·
paper
Diffusion Models for Recommendation
- DiffRec: Diffusion Recommender Model - Lin et al., RecSys (2023) · DDPM applied to CF; strong on cold/sparse data ·
paper - CF-Diff: Collaborative Filtering with Diffusion Models (2023) · multi-step denoising as personalization ·
paper
Part 13 - Exploration, Bandits & Reinforcement Learning
- Contextual Bandits for Personalized News Article Recommendation (LinUCB) - Li et al., Yahoo! (2010) ·
paper - Deep Bayesian Bandits Showdown - Riquelme et al. (2018) · benchmark of neural contextual bandits ·
paper - Top-K Off-Policy Correction for a REINFORCE Recommender System - Chen et al., Google (2019) · policy gradient with off-policy correction at YouTube ·
paper - Reinforcement Learning for Slate-based Recommender Systems (SlateQ) - Ie et al., Google (2019) · Q-learning for ordered list recommendations ·
paper - RecSim: A Configurable Simulation Platform for Recommender Systems - Google (2019) ·
paper - Unbiased Offline Evaluation of Contextual-Bandit-Based News Recommendation - Li et al. (2010) · replay method for offline bandit eval ·
paper
Part 14 - Causal Inference & Debiasing
- Recommendations as Treatments: Debiasing Learning and Evaluation - Schnabel et al. (2016) · IPS for unbiased recommendation training ·
paper - Unbiased Learning to Rank with Unbiased Propensity Estimation - Joachims et al. (2018) · position bias correction via dual learning ·
paper - CausE: Towards Causal Recommendations - Bonner & Vasile (2018) · counterfactual thinking for RecSys ·
paper - Deconfounded Recommendation for Alleviating Bias Amplification (2021) · causal graph + PD learning for popularity debiasing ·
paper - Disentangling User Interest and Conformity for Recommendation (2021) · DICE: causal embedding separating interest from conformity ·
paper - Popularity Bias in Recommender Systems - Abdollahpouri et al. (2021) · taxonomy of popularity bias types and effects ·
paper - A Survey on the Fairness of Recommender Systems (2022) · bias types, debiasing methods, fairness metrics ·
survey
Part 15 - Evaluation Methodology
- Evaluating Collaborative Filtering Recommender Systems - Herlocker et al. (2004) · canonical evaluation methodology paper ·
paper - Are We Really Making Much Progress? A Worrying Analysis of Recent Neural Recommendation Approaches - Dacrema et al. (2019) · reproducibility crisis: simple baselines match complex models ·
paper - Normalized Discounted Cumulative Gain (NDCG) for Ranking Evaluation - Järvelin & Kekäläinen (2002) ·
paper - Beyond Accuracy: Evaluating Recommender Systems by Coverage and Serendipity - Ge et al. (2010) · diversity + serendipity metrics ·
paper - A Closer Look at the Evaluation of Recommender Systems (2020) · sampled-based NDCG is heavily biased ·
paper - Leave-One-Out vs. Temporal Split Evaluation for Sequential RecSys (2020) · critiques common evaluation flaws in sequential rec ·
paper - Offline Evaluation of Recommender Systems: A Practitioner's View (2022) ·
paper - Netflix Tech Blog: A/B Testing Intuition Busters - Netflix ·
blog - RecBole: A Unified, Comprehensive and Efficient Recommendation Library (2020) ·
paper - Reclist: Beyond-accuracy Evaluation of RecSys - Chia et al. (2022) · behavioral testing framework for rec models ·
paper
Part 16 - Datasets
- MovieLens: Stable Benchmark Dataset - Harper & Konstan (2015) · 1M / 10M / 25M versions; starting point for most papers ·
dataset - Amazon Product Reviews (2018) - Ni et al. (2019) · 233M reviews across 29 categories; widely used for sequential rec ·
dataset - Yelp Open Dataset - local business reviews + metadata ·
dataset - MIND: Microsoft News Dataset - Wu et al. (2020) · 1M users, 160K news articles, click labels ·
dataset - KuaiRec: A Fully-Observed Dataset for Recommender Systems - Gao et al. (2022) · near-complete observation matrix; nearly unbiased evaluation ·
dataset - KuaiRand: An Unbiased Sequential Recommendation Dataset (2022) · random exploration policy logging for unbiased eval ·
dataset - Taobao User Behavior Dataset - Alibaba · 100M user behavior records ·
dataset - MSD: Million Song Dataset - Bertin-Mahieux et al. (2011) · implicit listening history for music rec ·
dataset - RecSys Challenge Datasets - ACM RecSys (annual) ·
dataset - LastFM Dataset - HetRec 2011 · music listening history ·
dataset
Part 17 - Surveys & Textbooks
- Recommender Systems Handbook (3rd ed.) - Ricci, Rokach & Shapira (2022) · comprehensive reference ·
book - Practical Recommender Systems - Falk, Manning (2019) · practitioner-focused ·
book - Deep Learning based Recommender System: A Survey and New Perspectives - Zhang et al. (2019) ·
survey - Graph Neural Networks in Recommender Systems: A Survey - Wu et al. (2020) ·
survey - Self-supervised Learning for Recommender Systems: A Survey - Yu et al. (2022) ·
survey - A Survey on Knowledge Graph-Based Recommender Systems - Guo et al. (2020) ·
survey - A Survey on Large Language Models for Recommendation - Wu et al. (2023) ·
survey - Toward the Next Generation of Recommender Systems - Adomavicius & Tuzhilin (2005) · foundational taxonomy survey ·
survey - Google ML Crash Course: Recommendation Systems · hands-on collaborative filtering, candidate generation, retrieval ·
website - RecSys Podcast (recspod.com) · practitioner interviews on production rec ·
website - How Can Recommender Systems Benefit from Large Language Models: A Survey - Lin et al. (2023/2024) · taxonomy of LLM roles: encoder, ranker, generator, agent ·
survey - A Survey on Diffusion Models for Recommendation (2024) · covers DiffRec, CF-Diff, and generation-based rec approaches ·
survey