<p>Point cloud quality assessment (PCQA) has become increasingly important due to its widespread application in fields such as virtual reality and autonomous driving, where accurate and reliable evaluation of point cloud data is critical for ensuring the fidelity and performance of these applications. However, obtaining distortion information from point cloud data is challenging, and the scarcity of labeled datasets adds to the difficulties in PCQA. In this paper, we propose a novel method that extends the concept of point cloud self-masking reconstruction and contrastive learning to the PCQA task. Specifically, this method involves preprocessing followed by two stages: pre-training and fine-tuning. The preprocessing stage partitions the point cloud into multiple patches to represent local distortions and serializes them according to their internal spatial proximity. In the pre-training stage, the framework comprises two distinct branches: a color point cloud reconstruction branch and a contrastive learning branch, both of which utilize a pre-trained CLIP encoder. The reconstruction branch improves texture feature representation, while the contrastive learning branch enhances distortion-wise feature representation. In the fine-tuning stage, we retain the reconstruction branch and adapt the CLIP encoder to the downstream PCQA task by unfreezing its parameters, which boosts overall performance. Experimental results show that our method outperforms the state-of-the-art no-reference PCQA method.</p>