Research map / Large language models / AI and society
Building Higher-Order Abstractions from the Components of Recommender Systems
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Presents MAB2REC, a modular open-source framework that integrates independent libraries (SELECTIVE, SEQ2PAT, TEXTWISER, MABWISER, JURITY) into a bandit-based recommender toolchain, with deployment and governance best practices.
Problem
Building recommender systems requires complex, monolithic machinery spanning data preparation, modeling, evaluation and deployment, which demands expert skills and engineering effort and is not transferable to other tasks or accessible to non-tech companies.
Approach
Decompose recommender systems into standalone, self-contained open-source components (feature selection, text featurization, sequential pattern mining, contextual bandits, fairness/performance evaluation) that are reusable beyond recommenders and integrate them into a unified bandit-based recommender toolchain.
Key results
- MAB2REC integrates SELECTIVE, SEQ2PAT, TEXTWISER, MABWISER and JURITY into one toolchain; the libraries surpassed 300K+ downloads in aggregate.
- TEXTWISER hosts more than 25 text embeddings with over 100 pre-trained models; MABWISER offers context-free, non-parametric and parametric MAB algorithms.
- The paper reports no new controlled benchmark results; evidence is deployment anecdotes (article recommendation, EaSe embeddings service).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Combine frequent and deep SEQ2PAT patterns | Models using SEQ2PAT patterns · E-commerce shopping intent prediction and intrusion detection | model performance | Helped |
| Use SEQ2PAT patterns as model features | Models built on SEQ2PAT patterns · E-commerce shopping intent prediction and intrusion detection | model performance | No clear effect |
Surprising findings
- A non-deterministic behavior of the LinTS bandit policy went unnoticed for 10+ years in the community, and the authors provided a fix and proof of correctness.
Limitations
- No new quantitative experiments comparing MAB2REC against other recommender frameworks.
- Evidence for deployment and adoption is anecdotal and largely self-reported from within the enterprise.
Experimental setup
Metrics: CTR IPS DR MAP NDCG inter-list diversity intra-list diversity coverage serendipity disparate impact
Baselines: Torch Recommenders TensorFlow Recommenders Microsoft Recommenders Merlin Recommenders Vowpal Wabbit Yelp MOE Contextual R GRU4Rec BERT4Rec Caser Transformer4Rec
Builds on: LinTS UCB1 Constraint-based sequential pattern mining Disparate Impact Statistical Parity
Closest papers by meaning
- AgentRecBench: Benchmarking LLM Agent-based Personalized Recommender Systems (NeurIPS 2025 · similarity 0.88)
- RecFlow: An Industrial Full Flow Recommendation Dataset (ICLR 2025 · similarity 0.86)
- Self-EvolveRec: Self-Evolving Recommender Systems with LLM-based Directional Feedback (NeurIPS 2026 · similarity 0.85)
- ORBIT - Open Recommendation Benchmark for Reproducible Research with Hidden Tests (NeurIPS 2025 · similarity 0.85)
- Recommender Ecosystems: A Mechanism Design Perspective on Holistic Modeling and Optimization (AAAI 2024 · similarity 0.84)
- Would You Like Your Data to Be Trained? A User Controllable Recommendation Framework (AAAI 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.