This paper has proposed a novel method to reduce the multi-keyword query traffic in Kademlia-like Peer-to-peer(P2P) networks by optimizing the Bloom filter settings. We build some models to estimate the communication cost, the union set size, and the loss rate of performing union and intersection operations. We implement a Kademlia-like system and generate a group of datasets.We use one part of the datasets to obtain the functions how to compute the optimal parameters and use another part of datasets to verify our method. Each query can determine the optimal settings of Bloom filter with no extra configuration. Our simulation experimental results show that with optimal Bloom filters settings, we can greatly reduce the communication cost under an acceptable loss rate. This paper has proposed a novel method to reduce the multi-keyword query traffic in Kademlia-like Peer-to-Peer (P2P) networks by optimizing the Bloom filter settings. We build some models to estimate the communication cost, the union set size, and the loss rate of performing union and intersection operations. We implement one Kademlia-like system and generate a group of datasets. We use one part of the datasets to obtain the functions how to compute the optimal parameters and use another part of datasets to verify our method. Each simulation can determine the optimal settings of Bloom filter with no extra configuration. Our simulation experimental results show that with optimal Bloom filters settings, we can greatly reduce the communication cost under an acceptable loss rate.