Many people use social networking services (SNSs) to easily access various news. There are numerous ways to obtain and share "fake news," which are news carrying false information. To address fake news, several studies have been conducted for detecting fake news by using SNS-extracted features. In this study, we attempt to use temporal features generated from SNS posts by using a point process algorithm to identify fake news from real news. Temporal features in fake news detection have the advantage of robustness over existing features because it has minimal dependence on fake news propagators. Further, we propose a novel multi-modal attention-based method, which includes linguistic and user features alongside temporal features, for detecting fake news from SNS posts. Results obtained from three public datasets indicate that the proposed model achieves better performance compared to existing methods and demonstrate the effectiveness of temporal features for fake news detection.