This paper proposes an inductive cognitive diagnosis model (ICDM) for fast new students' mastery levels inference in web-based online intelligent education systems (WOIESs). ICDM is designed to address the challenge of efficiently inferring mastery levels for new students in open learning environments without requiring retraining. The model introduces a novel student-centered graph (SCG) to represent students, exercises, and concepts, enabling the inference of mastery levels through aggregated information from neighbors in the graph. This approach shifts the task from finding student-specific embeddings to finding suitable representations for different node types in the graph, which is more efficient and avoids the need for retraining. The ICDM consists of a construction-aggregation-generation-transformation (CAGT) process to learn the final representations of students, exercises, and concepts. These representations are then used to predict students' performance on exercises through a global-level interaction function (GLIF). Extensive experiments on real-world datasets demonstrate that ICDM is significantly faster than existing transductive cognitive diagnosis methods while maintaining competitive inference performance for new students. The model's effectiveness is validated through various experiments, including comparisons with state-of-the-art methods and baselines. Results show that ICDM outperforms other methods in both transductive and inductive scenarios, particularly in terms of prediction accuracy and interpretability. Additionally, ICDM's ability to provide immediate feedback without retraining makes it highly suitable for WOIESs, where rapid response to student performance is crucial. The model's efficiency and effectiveness in handling large-scale data and diverse student interactions further highlight its potential for real-world applications in online education.This paper proposes an inductive cognitive diagnosis model (ICDM) for fast new students' mastery levels inference in web-based online intelligent education systems (WOIESs). ICDM is designed to address the challenge of efficiently inferring mastery levels for new students in open learning environments without requiring retraining. The model introduces a novel student-centered graph (SCG) to represent students, exercises, and concepts, enabling the inference of mastery levels through aggregated information from neighbors in the graph. This approach shifts the task from finding student-specific embeddings to finding suitable representations for different node types in the graph, which is more efficient and avoids the need for retraining. The ICDM consists of a construction-aggregation-generation-transformation (CAGT) process to learn the final representations of students, exercises, and concepts. These representations are then used to predict students' performance on exercises through a global-level interaction function (GLIF). Extensive experiments on real-world datasets demonstrate that ICDM is significantly faster than existing transductive cognitive diagnosis methods while maintaining competitive inference performance for new students. The model's effectiveness is validated through various experiments, including comparisons with state-of-the-art methods and baselines. Results show that ICDM outperforms other methods in both transductive and inductive scenarios, particularly in terms of prediction accuracy and interpretability. Additionally, ICDM's ability to provide immediate feedback without retraining makes it highly suitable for WOIESs, where rapid response to student performance is crucial. The model's efficiency and effectiveness in handling large-scale data and diverse student interactions further highlight its potential for real-world applications in online education.