Abstract: The purpose of knowledge tracking is to predict learners' future answers and evaluate their knowledge mastery based on their historical answer records. Most existing deep knowledge tracking methods fail to effectively utilize the diverse behavioral information in datasets, and have difficulties in distinguishing the response benefits of different learners. To address these problems, a deep knowledge tracing model that integrates learners' knowledge absorption capacity (Knowledge Absorptive Capacity Deep Knowledge Tracing, KADKT) is proposed. Firstly, in order to solve the problem of insufficient utilization of behavioral information, the extracted behavioral information is divided into explicit behavior and implicit behavior, and the difficulty information and learners' answer level (ave-rage answer time and average number of attempts) of the question are calculated by explicit behavior. The implicit behavior represents the situation that the learner has learned a certain knowledge in history; secondly, in order to calculate the learner's answer return individually, a knowledge absorptive capacity modeling method that integrates the difficulty and the average answer situation is proposed; thirdly, the learner's answer return is calculated from the knowledge point level and the interaction level respectively to capture the independent contribution of the know-ledge point to the answer return; finally, a forgetting module is designed to simulate the learner's knowledge forgetting and comprehensively update the change of the learner's knowledge state. Comparative experiments are conducted between the KADKT model and six benchmark models DKT, DKVMN, SAKT, AKT, LPKT, and LBKT on three public datasets(ASSISTments2009, ASSISTments2017, and Junyi Academy). The results of the comparative experiments showed that the KADKT model outperformed other benchmark models in terms of area under the curve (AUC), accuracy (ACC), and root mean square error (RMSE) across the three datasets. Specifically, on the ASSISTments2009 dataset, compared with the benchmark models, the KADKT model achieved an increase of 2.6% to 10.61% in AUC, 3.28% to 7.93% in ACC, and a decrease of 2.24% to 6.52% in RMSE. In addition, the ablation experiment verified the validity of each module of the KADKT model, and the visual analysis proved that the KADKT model could effectively track the changes of learners' knowledge status.