KVQ is a large-scale database for quality assessment of short-form videos, containing 600 user-uploaded and 3600 processed videos. It was created to address the challenges of quality assessment in short-form user-generated content (S-UGC) videos, such as ambiguous content and complex distortions. The database includes absolute quality scores and partial rankings for indistinguishable samples, provided by professional researchers. Based on this database, the first short-form video quality evaluator, KSVQE, was proposed. KSVQE uses large vision-language models (CLIP) for content understanding and a distortion understanding module to distinguish distortions. It also incorporates a quality-aware region selection module (QRS) and content-adaptive modulation (CaM) to enhance content understanding and a distortion-aware modulation (DaM) to improve distortion understanding. Experimental results show that KSVQE outperforms existing methods on the KVQ database and popular VQA databases. The database and evaluator have broad applicability for UGC-VQA tasks. The paper also presents an ablation study of the components of KSVQE, demonstrating their effectiveness in improving performance. The results show that KSVQE achieves state-of-the-art performance on the KVQ dataset and other benchmark datasets.KVQ is a large-scale database for quality assessment of short-form videos, containing 600 user-uploaded and 3600 processed videos. It was created to address the challenges of quality assessment in short-form user-generated content (S-UGC) videos, such as ambiguous content and complex distortions. The database includes absolute quality scores and partial rankings for indistinguishable samples, provided by professional researchers. Based on this database, the first short-form video quality evaluator, KSVQE, was proposed. KSVQE uses large vision-language models (CLIP) for content understanding and a distortion understanding module to distinguish distortions. It also incorporates a quality-aware region selection module (QRS) and content-adaptive modulation (CaM) to enhance content understanding and a distortion-aware modulation (DaM) to improve distortion understanding. Experimental results show that KSVQE outperforms existing methods on the KVQ database and popular VQA databases. The database and evaluator have broad applicability for UGC-VQA tasks. The paper also presents an ablation study of the components of KSVQE, demonstrating their effectiveness in improving performance. The results show that KSVQE achieves state-of-the-art performance on the KVQ dataset and other benchmark datasets.