Arterial blood pressure (ABP) monitoring plays a vital role in the prevention of cardiovascular diseases. However, conventional cuff-based devices are unsuitable for continuous monitoring due to their lack of portability and comfort. Photoplethysmography (PPG) sensors have been explored in recent studies for ABP measurement, nevertheless, achieving a balance between accuracy and user burden remains a difficulty. This study proposes a deep learning-based ABP monitoring system SimilarBP that combines the strengths of personalized and user-independent models, aiming to achieve high ABP measurement accuracy with few-shot personal data. SimilarBP pre-trains a user-independent model on a large dataset, then fine-tunes it using few-shot personal data, along with similar samples from the large dataset. Challenges include identifying similar samples and reducing the impact of incorrect ABP labels. To address these challenges, this study proposes an individualized contrastive learning model (ICLM) for identifying similar samples and an ABP label correction algorithm (ALCA) for correcting ABP labels. SimilarBP is validated on a public dataset and a real-world volunteer dataset. Evaluation results demonstrate that SimilarBP meets the AAMI standard.