This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
Beyond Static Testbeds: An Interaction-Centric Agent Simulation Platform for Dynamic Recommender Systems
Created by
Haebom
Author
Song Jin, Juntian Zhang, Yuhan Liu, Xun Zhang, Yufei Zhang, Guojun Yin, Fei Jiang, Wei Lin, Rui Yan
Outline
RecInter is a novel agent-based simulation platform for evaluating and improving recommender systems. To overcome the resource-intensive nature of traditional A/B testing and the limitations of offline methods, it provides a dynamic interaction mechanism where user actions change item attributes in real time and merchant agents respond accordingly. RecInter ensures a high level of simulation accuracy through multidimensional user profiling, advanced agent architecture, and fine-tuned LLM with enhanced interaction data using Chain-of-Thought (CoT) methods. This allows it to replicate phenomena such as brand loyalty and the Matthew Effect, establishing it as a reliable testbed for recommender system research.
Takeaways, Limitations
•
Takeaways:
◦
Providing a realistic simulation environment for recommender system research.
◦
Realistic system evolution simulation that reflects dynamic interactions between user behavior and the platform.
◦
Reproducing complex phenomena such as brand loyalty and the Matthew effect
◦
Suggesting the possibility of using it as an alternative to A/B testing.
•
Limitations:
◦
Lack of detailed information on specific platform performance, scalability, and efficiency.
◦
Lack of information on the use of LLM and the specific implementation of the CoT approach.
◦
Lack of practical application examples of the proposed platform for recommendation systems.
