The emergence of a large amount of pharmacological, genomic, and network knowledge data provides new challenges and opportunities for drug discovery and development. Identification of real small-molecule drug (SM)-miRNA associations is not only important in the development of effective drug repositioning but also crucial in providing a better understanding of the mechanisms by which small-molecule drugs achieve the purpose of treating diseases by regulating miRNA expression. However, challenges remain in accurately determining potential associations between small molecules and miRNAs using information from multiomics data. In this study, we adopted a novel framework called SMAJL to improve the prediction of small molecule-miRNA associations with joint learning. First, we use enhancing matrix completions to obtain the network knowledge of small molecule-miRNA associations. Then, we extract the information of small-molecule fingerprints and miRNA sequences into feature vectors to obtain small-molecule structure and miRNA sequence information. Finally, we incorporate small-molecule structure information, miRNA sequence data, and heterogeneous network knowledge into a joint learning model based on a Restricted Boltzmann Machine (RBM) to predict association scores. To validate the effectiveness of our method, the SMAJL model is compared with four state-of-the-art methods in terms of 5-fold cross-validation. The results demonstrate that the AUC and AUPRC of the SMAJL are obviously superior to those of other comparison methods. The SMAJL model also achieved great results in terms of robustness and case studies, further demonstrating its strong predictive power.