The fast development of social media fuels massive spreading of misinformation, which harm information security at an increasingly severe degree. It is urgent to achieve misinformation identification and early detection in social media. However, two main difficulties hinder the identification of misinformation. First, an event about a piece of suspicious news usually comprises massive microblog posts (hereinafter referred to as post), and it is hard to directly model the event with massive-volume posts. Second, information in social media is of high noise, i.e., most posts about an event have little contribution to misinformation identification. To resolve the difficulty of massive volume, we propose an Event2vec module  to learn distributed representations of events in social media. To overcome the difficulty of high noise, we mine significant posts via content and temporal co-attention, which learn importance weights for content and temporal information of events. In this paper, we propose an Attention-based Convolutional Approach for Misinformation Identification (ACAMI) model. The Event2vec module and the co-attention contribute to learning a good representation of an event. Then the Convolutional Neural Network (CNN) can flexibly extract key features scattered among an input sequence and shape high-level interactions among significant features, which help effectively identify misinformation and achieve practical early detection. Experimental results on two typical datasets validate the effectiveness of the ACAMI model on misinformation identification and early detection tasks.