<p>When conducting patent transactions, companies usually purchase a set of related patents, called patent portfolios. With the development of e-commerce, some online patent trading platforms have emerged. However, these platforms still face difficulties in identifying and implementing transactions, which can be alleviated by patent recommendation. In reality, not only can patents traded by a company at the same time be considered as a patent portfolio with temporal characteristics, but also due to the universal connection between patents owned by the company, all patents of the company can be considered as an overall patent portfolio. Therefore, our study proposes a deep learning-based patent portfolio recommendation model that considers both the overall patent portfolio and the sequential patent portfolio. In the model, we construct a patent representation system, introduce a Lite-Transformer to learn the characteristics of each patent in the overall patent portfolio, and then represent the sequential patent portfolio based on time division. We also use a sequence encoder to learn the relationship between the sequential patent portfolio, and a score predictor is used to make final recommendations. Our model outperforms several baselines in comparative experiments, and we demonstrated the differences in recommendation mechanisms between our model and the general patent recommendation model through a case study.</p>