Conduct a vary-k experiment based on the following configuration
- extending Ad-LLM to multiple k:
- currently: select the best one product and its injection position
- update: select the top-k products based on the disturbance minimization and the corresponding positions. (Note: if meet conflict, in sequential order to place the consecutive ads)
- solution: Ad-LLM (GIR-R)
- base LLM: doubao-1-5-lite-32k; judge LLM: gpt-4.1-mini
- embedding model: text-embedding-3-small
- dataset: LM-SYS-100
- setting of k: {2, 3, 4, 5}
Conduct a vary-k experiment based on the following configuration