To solve the problem of deadlock occurred in multi-issue negotiation, an optimization strategy is presented in this paper, which uses a learning method to estimate opponents' preferences of their negotiation issue, and gives full consideration to the issue dependencies. Considering the weighted value of the opponent, we adjust the reserved value of some issues accordingly, by which we can achieve the purpose of solving the deadlock problem quickly and efficiently, and the negotiation participants can get a reasonable solution at the same time. Experimental results prove the efficiency of the whole negotiation process.