A survey organizes LLM-based recommendation systems by autonomy level and distinguishes agent-assisted recommendation, the agent as recommender, and the agent as user simulator.
A survey of LLM-based agents in recommendation systems proposes a taxonomy organized by level of autonomy. Its aim is to help engineers distinguish design patterns.
The first paradigm is agent-assisted recommendation. The second treats the agent as the recommender itself. The third uses the agent as a user simulator.
When evaluating an architecture, identify the agent’s role and how much autonomy it receives. Then compare the options against your system’s requirements and constraints; do not assume one paradigm is universally better.
If you use AI to study or apply the survey, avoid submitting internal data, personal data, or confidential examples. Use public or anonymized material and follow your organization’s rules.