With the fast-changing development of emerging online media, it has be-come apparent that information on social networks is characterized by extensive, fast and timely spreading. The absence of effective detection methods and moni-toring means has led to a massive outbreak of rumors. Therefore, accurate detection and timely suppression of rumors in social networks is a vital task in maintaining social security and purifying public networks. Most existing work relies only on monotonous textual content and shallow semantic information, and lacks critical at-tention to and potential mining of user relationships. Such being the case, we can better improve these problems by employing attention mechanisms. In this paper, we proposea Multi-Attention Neural Interaction Network (MANIN) for rumor detection, which consists mainly of a self-attention-based BERT encoder, a post-comment co-attention mechanism, and a graph attention neural network for mining potential user interactions. We have conducted numerous experiments on real datasets and verified their validity, and the results show that the model proposed by us outperforms existing models with an accuracy rate of 81.6%.