In today’s world, communication or relationships across multiple domains can be represented by multi-channel networks, where entities can communicate at multiple layers. In the current community detection algorithms for multi-channel networks, the multi-level Memetic algorithm has shown excellent performance, but still ignores the existence of sparse networks. This article proposes a multilevel memetic community detection algorithm based on Jaccard mutation and multi neighbor search: MMCD-JM. Firstly, combining Jaccard similarity coefficient with genetic algorithm for mutation operation in global search can more effectively find neighboring nodes for mutation on sparse networks, avoiding performance degradation. Secondly, this article proposes a search strategy that adds search targets to multiple neighbors in community level local search to improve the accuracy of results and find good community partitions with high module values and redundancy values. Finally, extensive experiments were conducted on six real networks in multiple fields in the real world, and the results showed that our algorithm was more effective in identifying communities in multi-layer networks.