Back to Open-Source AgentsPrimary resources: [GitHub repository](https://github.com/LehengTHU/Agent4Rec), [SIGIR 2024 paper (arXiv)](https://arxiv.org/abs/2310.10108), [ACM Digital Library](https://dl.acm.org/doi/10.1145/3626772.3657844) Type: **Open-source simulator** for recommender systems using **LLM-based generative agents** Origin: Based on the SIGIR 2024 paper, *On Generative Agents in Recommendation* Scale: **1,000 agents** initialized from MovieLens-1M and extended with profiles, memory (including emotions/reflection), and an action policy Domains: **MovieLens-1M**, **Amazon-Book**, **Steam** Status: Research artifact (not a commercial product or hosted service) Study real-time behavior instead of static clicks to better understand **exposure → consumption → feedback loops** Safely test ranking policies and interventions offline Generate **synthetic yet faithful behavioral data** to augment sparse datasets Investigate **filter bubbles**, **popularity bias**, and **causal relationships** among item quality, exposure, and engagement Agents are initialized from real datasets (e.g., MovieLens-1M) and driven by **LLM prompts** for memory and decisions A **collaborative filtering baseline** provides recommendations; agents respond in a structured loop across pages Interaction log includes actions: **watch**, **rate**, **evaluate**, **exit**, and **interview** Designed for controlled experimentation without live users Learn more in the [GitHub repo](https://github.com/LehengTHU/Agent4Rec) and the [SIGIR 2024 paper](https://arxiv.org/abs/2310.10108) Interactive user simulation with memory and emotion modeling Cross-domain support (movies, books, games) for generalization tests Data augmentation via agent-generated ratings and interactions Tools for **causal analysis** of quality, popularity, exposure, and consumption Language/stack: **Python**, **PyTorch** Recommended versions: Python 3.9.12; torch 1.13.1+cu117 (note: issues reported with Python > 3.10 due to reckit) Typical steps: Install dependencies: `pip install -r requirements.txt` Build extensions from `recommenders/`: `python setup.py build_ext --inplace` Datasets: initialize with **MovieLens-1M**; experiments include **Amazon-Book** and **Steam** Full instructions in the [repository](https://github.com/LehengTHU/Agent4Rec) Recommender systems researchers and PhD students Academic labs studying **agentic RS**, **user simulation**, or **RL over feedback loops** R&D teams needing offline policy testing and **counterfactual/causal analysis** Benchmark and dataset curators building **agentic evaluation environments** Offline evaluation of ranking policies with **interactive** feedback Measuring and mitigating **filter bubbles** and exposure bias Generating synthetic labels/interactions to **reduce sparsity** Causal studies: disentangling **quality vs. popularity vs. exposure** Cross-domain stress testing (movies/books/games) Prototyping **agentic recommenders** with memory/emotion-driven actions Benefits Enables safe, repeatable policy experiments without live users Produces structured logs for **longitudinal analysis** Cited in surveys and repos, reinforcing research credibility: see [survey 1](https://arxiv.org/html/2503.05659v1), [survey 2](https://arxiv.org/html/2503.16734v1) Limitations LLM simulators aren’t ground truth; risk of **overfitting to simulation artifacts** Running large agent populations can be **compute/cost intensive** No hosted service or enterprise benchmarks (e.g., G2/Capterra); positioned as **research-only** Paper: [On Generative Agents in Recommendation (SIGIR 2024)](https://arxiv.org/abs/2310.10108) | [ACM DL entry](https://dl.acm.org/doi/10.1145/3626772.3657844) Authors: An Zhang, Yuxin Chen, Leheng Sheng, Xiang Wang, Tat-Seng Chua Directory listings (research context): [AI Agents List](https://aiagentslist.com/agent/agent4rec), [AI Agent Store](https://aiagentstore.ai/ai-agent/agent4rec), [AgentLocker](https://www.agentlocker.ai/agent/agent4rec) Not a commercial company or hosted product; no LinkedIn company page, G2, or Capterra listings Project is maintained as an **open-source research codebase**: [GitHub](https://github.com/LehengTHU/Agent4Rec) Keywords: generative agents, user simulation, recommender systems, LLM agents, filter bubbles, exposure bias, offline evaluation, data augmentation, SIGIR 2024, MovieLens-1M Related ecosystems: Python, PyTorch, collaborative filtering, agent-based modeling Paper PDF mirror: [SciSpace PDF](https://scispace.com/pdf/on-generative-agents-in-recommendation-3bxda56n9g.pdf) Additional directories: [AI Agents Directory](https://aiagents.directory/agent4rec/), [AI Agents Base](https://aiagentsbase.com/agents/agent4rec)
Funding
OSS
Agent4Rec: LLM-Driven User Simulation for Recommender Systems
Agent4Rec is an open-source research simulator that models interactive user behavior for recommender systems using 1,000 LLM-driven generative agents. Instead of static click logs, it simulates page-by-page interactions—watching, rating, evaluating, exiting, and interviews—allowing researchers to study longitudinal dynamics such as filter bubbles, exposure effects, and data augmentation.
What It Is
Why It Matters
How It Works
Core Capabilities
Setup and Environment
Who It’s For
High-Value Use Cases
Benefits and Limitations
For community perspectives, see discussions on multi-agent systems and complexity: [discussion 1](https://www.reddit.com/r/AI_Agents/comments/1j9bwl7/do_we_actually_need_multiagent_ai_systems/), [discussion 2](https://www.reddit.com/r/LangChain/comments/1izsw0u/which_framework_you_use_for_multiagents/), [discussion 3](https://www.reddit.com/r/LangChain/comments/1byz3lr/insights_and_learnings_from_building_a_complex/).
Research and Recognition
Availability and “Company” Status
SEO Quick Facts
Related Resources
If you need, we can extract README details on running simulations, config files, and evaluation scripts directly from the [repository](https://github.com/LehengTHU/Agent4Rec).