Patent analysis requires significant effort and expertise in domain-specific technologies, information retrieval, and business intelligence. Patent classification is one of the essential steps for managing and maintaining patent documents. Hence, automating this expensive and laborious task is essential for assisting domain experts in managing patent documents, facilitating reliable search, retrieval, and further patent analysis tasks. However, the text used in patent documents is not always written in a way to efficiently convey knowledge. Therefore, the conventional text classification methods fail to process patent text. This thesis focuses on investigating and utilizing deep learning-based techniques, especially natural language understanding models, for automatically extracting features from patent text and improving the performance of multi-label patent classification task. Moreover, a novel classification model is proposed for the hierarchical multi-label patent classification problem. Experiments conducted on real data sets show a new state-of-the-art classification performance.