| Personalized learning aims to tailor learning resources and learning plans to learners’ differences and cognitive levels to compensate for the deficiencies in the existing knowledge structure and increase learning efficiency.More and more learning process data has been amassed in the education area in recent years,with the progressive popularization of the application of big data technologies in numerous fields,allowing for tailored learning based on the learner’s knowledge structure.However,how to diagnose learners’ knowledge status(i.e.,cognitive diagnosis)based on learning process data and support learners’ personalized and targeted learning has become a research hotspot in the field of education.Cognitive diagnosis has a more than 30-year history,with more than 100 models established,which solves the problem of learners’ knowledge structure diagnosis to some extent,but this field is still faced with challenges such as the complexity of cognitive processes and the variability of learning states.Therefore,this paper systematically proposes cognitive diagnostic approaches for learners’ cognitive process and its application in education.Specifically,aiming at the complexity of learners’ cognitive process,this paper proposes a hierarchical cognitive diagnostic model integrating multidimensional features to investigate the influence mechanism among multi-layer cognitive factors.Under the guidance of the hierarchical cognitive diagnosis model,aiming at the variability of learning states in single and multiple assessment scenarios,we propose a cognitive diagnosis model based on online interaction features with response time,a cognitive diagnosis approach based on learning and forgetting factors,and a dynamic cognitive diagnosis approach based on timeline sequence data in the assessment process.First,we propose a hierarchical cognitive diagnostic model integrating multidimensional features.To address the complexity of learners’ cognitive processes,a hierarchical cognitive diagnosis model based on the nature of the phenomenon is proposed,with a two-layer structure based on external characteristics and cognitive laws.To begin,based on pedagogical theory,the external characteristics of learners’ interactions with items for the assessment process are thoroughly analyzed,and they are divided into four aspects:knowledge characteristics,interaction characteristics,behavior characteristics,and timing characteristics.Besides,combined with learning theory,we explore complex correlations between multidimensional external features and cognitive laws and propose a hierarchical cognitive model based on cognitive laws to ultimately realize the prediction of learners’ future performance and the diagnosis of knowledge mastery.Second,we propose a cognitive diagnosis model based on online interaction features with response time(RT-CDM).To address the one-sided diagnostic results problem produced by the restricted interactive features in traditional cognitive diagnostics,we introduce the interactive features(i.e.response time),which investigates the impact of speed on the learner’s knowledge state.The learner’s response time on each item can indicate the learner’s knowledge mastery status through a continuous latent feature function on the accuracy side and a response time function on the speed side based on the"speed-accuracy" exchange criterion.It adopts the Markov Chain Monte Carlo(MCMC)algorithm for parameter estimation.To evaluate the performance of the method,we apply it to the PISA 2015 math dataset based on computer tests and conduct simulation studies under ideal conditions.The results of the experiments show that adding response time to the method improves its stability and accuracy.Third,we propose a dynamic cognitive diagnosis approach based on learning and forgetting factors(CF-DKD).Aiming at the inconsistency and instability of knowledge states caused by cognitive laws in temporal dynamic cognitive diagnosis,we propose a dynamic cognitive diagnosis combining learners’ cognitive laws based on the hierarchical cognitive diagnosis model.Considering characteristics of unequal time interval and sparse data in the real scene of Chinese assessment,CF-DKD combines the internal cognitive laws(i.e.learning and forgetting)with the key-value memory network to store potential item information and capture long-term time characteristics by wakening or enhancing knowledge memory through two gate mechanisms.Experiments and visualization experiments on four real datasets in K-12 education demonstrate that the method can handle time-series data efficiently and that its prediction results are better and more stable than existing baseline models.Fourth,we propose a dynamic cognitive diagnosis approach based on timeline sequence data in the assessment process(TDCD).Aiming at the problem of unstable diagnosis caused by the lack of complete modeling on time series data in traditional dynamic cognitive diagnosis methods,this study focuses on the learners’ assessment process.TDCD divides the external features into inner-time cells and inter-time cells and uses a flexible attention mechanism to simulate the learner’s evolution of knowledge state according to internal cognitive mechanisms such as speed,forgetting,and learning,realizing dynamic diagnosis of knowledge state.Finally,a significant number of experiments on two real large-scale datasets verify the model’s good performance,and the influence of cognitive laws on the learner’s knowledge state is verified from different perspectives.Fifth,we propose a learning early warning dashboard system based on cognitive diagnosis(LAD).To explore how the approach of cognitive diagnosis can be applied in the education system and its impact on learners,a learning early warning dashboard system based on cognitive diagnosis is designed and experiments are carried out in a real teaching environment,based on "feedback-correction"conceptual model.Finally,taking the eighth-grade mathematics course as the research object,the effectiveness of the proposed learning early warning dashboard system is assessed by contrasting the learning performance of the experimental class using the learning early warning dashboard and the control class using the traditional unit assessment learning model.Experimental results show that:(1)The learning early warning dashboard system based on cognitive diagnostic can greatly increase learners’ learning performance,particularly for pupils with poor academic levels.(2)Learning dashboards can help learners greatly improve their learning strategies.In conclusion,dynamic cognitive diagnostic models integrating multidimensional features span multiple fields such as educational measurement,computer,and psychology,which promote personalized learning.This study introduces cognitive rules to model the learners’ assessment process integrating multidimensional features.We also research a variety of cognitive diagnostic approaches to achieve accurate diagnosis of learners’ knowledge status. |
