AI-generated image quality assessment (AIGIQA), which aims to evaluate the perceptual quality of AI-generated images (AIGIs), has emerged as a novel topic in the field of computer vision. Most existing studies primarily focus on evaluating the quality of text-to-image AIGIs and adopt no-reference image quality assessment (NR-IQA) methods due to the lack of reference images. However, conventional NR-IQA approaches often rely solely on the target image, neglecting valuable comparative information from other images in the databases, which may restrict model performance. To overcome this limitation, we first introduce the contrastive regression (CR) framework to the AIGIQA task, enabling inter-image comparisons within the database rather than isolated assessments. Moreover, to better align with AIGI evaluation requirements, we propose a text-image encoder-based contrastive regression (TIECR) framework that jointly leverages both AIGIs and their corresponding text prompts for regression. We evaluate effectiveness of TIECR framework through comprehensive experiments conducted on AGIQA-1K, AGIQA-3K, and AIGCIQA2023 databases. The experimental results indicate that the proposed TIECR framework demonstrates superior performance over baselines, showing remarkable accuracy in assessing text-image correspondence. Code are released on https://github.com/jiquan123/PSCR.