Abstract: Predicting learners’ future performance based on their historical learning interactions is a fundamental task in knowledge tracing. Existing knowledge tracing models remain limited in modeling the interaction between multi-source static semantic information and dynamic behavioral sequences, and their prediction processes often lack interpretability, which restricts their reliability in real educational applications. To address these issues, we propose a knowledge tracing model based on conditional feature modulation with explicit interpretability (CFMKT). The model builds upon learners’ sequential response data and incorporates multiple types of static semantic features, including item difficulty, knowledge concept difficulty, and a knowledge relation graph. An improved feature-wise linear modulation (FiLM) mechanism is introduced to conditionally modulate the dynamic sequence representations at the feature level, enabling deep interaction between static semantics and learners’ temporal behaviors. Moreover, the model embeds interpretable intermediate variables into both the modulation and prediction processes, allowing prediction outcomes to be traced through historical evidence, static semantic evidence, and their adaptive weighting. This design provides structural transparency and enhances the interpretability of the knowledge tracing process. Experiments conducted on three real-world educational datasets demonstrate that the proposed model consistently outperforms several state-of-the-art methods in terms of both prediction accuracy and interpretability, validating its effectiveness in multi-source semantic interaction modeling and explainable learner state estimation.