Multi-agent KTO: Reinforcing Strategic Interactions of Large Language Model in Language Game