CN-122021721-A - Knowledge tracking method and system for decoupling cognitive state

CN122021721ACN 122021721 ACN122021721 ACN 122021721ACN-122021721-A

Abstract

The invention discloses a knowledge tracking method and a knowledge tracking system for a decoupling cognitive state, which relate to the technical field of knowledge tracking, wherein the knowledge tracking system for the decoupling cognitive state mainly comprises a problem and concept embedding representation module, a rule and concept embedding representation module and a rule and concept embedding representation module, wherein the problem and concept embedding representation module is used for constructing a problem-concept heterogeneous relationship graph and a triplet relationship thereof, learning the embedding representation of the problem and the concept according to the triplet relationship to obtain a problem vector and a concept vector, and obtaining basic interaction embedding according to interaction; the system comprises a cognitive state decoupling module, a cognitive state tracking module and a prediction probability obtaining module, wherein the cognitive state decoupling module is used for constructing a fluctuation cognitive state and a stable cognitive state, and the cognitive state tracking module is used for fusing the fluctuation cognitive state and the stable cognitive state and carrying out attenuation attention tracking to obtain the prediction probability. By implementing the knowledge tracking method of the decoupling cognitive state, provided by the invention, the prediction performance, the interpretability, the robustness and the applicability of the knowledge tracking model can be improved.

Inventors

XIAO KUI
HAN JINYING
ZHANG MIAO
WANG CHENG
HUANG ZHIFANG
LI ZHIFEI
LV XIAOPAN

Assignees

湖北大学

Dates

Publication Date: 20260512
Application Date: 20260407

Claims (10)

1. The knowledge tracking system for decoupling the cognitive state is characterized by comprising a problem and concept embedding representation module, a cognitive state decoupling module and a cognitive state tracking module; The problem and concept embedding representation module is used for constructing a problem-concept heterogeneous relation diagram and a triplet relation thereof, learning the embedding representation of the problem and concept according to the triplet relation, obtaining a problem vector and a concept vector, and obtaining basic interaction embedding according to interaction; The cognitive state decoupling module is used for constructing a fluctuation cognitive state and a stable cognitive state; the cognitive state tracking module is used for fusing the fluctuation cognitive state and the stable cognitive state and carrying out attenuation attention tracking to obtain prediction probability.
2. The knowledge tracking system of decoupled cognitive states of claim 1, wherein the problem-concept heterogeneous relationship graph comprises a set of nodes and a set of edges.
3. The knowledge tracking system for decoupling cognitive states of claim 1, wherein said method for obtaining problem vectors and concept vectors comprises training said triplet relationships using a knowledge graph embedding model to obtain problem vectors and concept vectors.
4. The knowledge tracking system for decoupling cognitive states of claim 1, wherein the method for obtaining problem vectors and concept vectors further comprises extracting concept text vectors using a pre-trained language model and fusing the concept text vectors with the concept vectors to obtain final concept vectors.
5. The knowledge tracking system of decoupling cognitive states of claim 1, wherein the method of generating a basic interaction embedment comprises, for each interaction, concatenating the problem vector and the concept vector with a coded vector reflecting the correct or incorrect, forming an interaction representation of the current time step, resulting in a basic interaction embedment.
6. The knowledge tracking system of decoupled cognitive states of claim 1, wherein the cognitive state decoupling module is further configured to construct a fluctuating cognitive feature vector, the constructing comprising: calculating subjective experience scalar, mental load scalar and mental effort scalar based on the answer result and the problem difficulty for each time step; Mapping the subjective experience scalar, the mental load scalar and the mental effort scalar into a subjective experience vector, a mental load vector and a mental effort vector by using a multi-layer perceptron; And splicing the subjective experience vector, the mental load vector and the mental effort vector with the basic interactive embedding to obtain the fluctuation cognition feature vector.
7. The knowledge tracking system of decoupled cognitive states of claim 1, wherein the process of constructing the fluctuating cognitive states comprises: constructing a directed concept conversion graph according to a learner concept sequence, wherein nodes of the directed concept conversion graph are concepts, and edges of the directed concept conversion graph represent concept switching of adjacent answers; obtaining an in-edge adjacent matrix and an out-edge adjacent matrix according to the constructed directed concept conversion diagram; and converting neighborhood information by utilizing the graph neural network aggregation concept according to the fluctuation cognition feature vector, the in-edge adjacent matrix and the out-edge adjacent matrix to obtain the fluctuation cognition state of the learner.
8. The knowledge tracking system of decoupled cognitive states of claim 1, wherein the process of constructing the stable cognitive states comprises: according to the basic interaction embedding, a sequence model is utilized to obtain a stable cognitive state of the learner evolving along with time; and dispersing and embedding the problem difficulty into a plurality of grades, estimating the deviation of the current problem difficulty relative to the average problem difficulty in a given stable state, and subtracting the deviation from the original stable cognitive state to obtain the unbiased stable cognitive state.
9. The knowledge tracking system of decoupled cognitive states of claim 1, wherein the cognitive state tracking module is specifically configured to: Fusing the fluctuation cognitive state and the stable cognitive state, introducing a position distance-based attenuation attention mechanism to the history interaction, and giving higher weight to the adjacent interaction to obtain the final cognitive state of the student; And combining the final cognitive state of the student with the target problem and the concept vector, and outputting correct answer probability at the next moment through linear mapping and Sigmoid function.
10. A knowledge tracking method for decoupling cognitive states, characterized in that a knowledge tracking system for decoupling cognitive states according to any one of claims 1-9 is used for knowledge tracking.

Description

Knowledge tracking method and system for decoupling cognitive state Technical Field The invention relates to the technical field of knowledge tracking, in particular to a knowledge tracking method and a knowledge tracking system for decoupling cognitive states. Background An online learning platform (such as an intelligent coaching system, a question bank platform, a MOOC, etc.) records the interaction sequence (question ID/knowledge point ID/answer correctness/answer sequence, etc.) of the learner and the questions. The goal of knowledge tracking (Knowledge Tracing, KT) is to predict the correct probability of answering a target problem at the next moment according to the learner's historical interaction sequence, thereby serving personalized recommendation, weak knowledge diagnosis and teaching intervention. The prior knowledge tracking method mainly comprises (1) a traditional probability model (such as Bayesian knowledge tracking), carrying out state transition on knowledge mastering through preset cognitive hypothesis, (2) a depth sequence model (such as RNN/GRU/LSTM-based depth knowledge tracking), encoding the 'knowledge state' of a learner end to end by using hidden state vectors, (3) improving the expression capacity of a long sequence through global dependency modeling based on a attention/transducer model, (4) a small amount of work starts focusing on cognitive fluctuation or uncertainty, but most of work still is characterized in a single hidden state or an indirect mode, and the fluctuation cognitive state is not modeled in a display mode. Despite the significant progress made by prior knowledge tracking methods in predicting student response performance, two key bottlenecks remain. (1) insufficient interpretability due to cognitive state coupling. The existing model generally models the cognitive state of students as a single hidden variable, so that short-term cognitive fluctuation and long-term knowledge accumulation in the learning process of the students are difficult to distinguish, and misjudgment on the true capability of the students is more likely to be caused. (2) knowledge assessment bias caused by random factor interference. In addition, the sequence dependence of the learning path and the difference of the difficulty of the exercises can also interfere with the capability assessment, so that the model can bias the true capability of the students. In summary, a knowledge tracking method capable of explicitly decoupling short-term cognitive fluctuations from long-term knowledge accumulation while mitigating problem difficulty bias is needed to improve prediction performance and interpretability. In the prior art, cognitive fluctuation is considered to a certain extent, but most of the cognitive fluctuation does not have intrinsic factors influencing the fluctuation cognitive state, such as subjective experience, mental load, psychological effort and other cognitive characteristics of students which dynamically change in the answering process, so that the dynamic fluctuation of the cognitive state in different learning stages can not be fully captured. Disclosure of Invention The invention aims to provide a knowledge tracking method for decoupling cognitive states, which can improve the prediction performance, the interpretability, the robustness and the applicability of a knowledge tracking model. The invention provides a knowledge tracking system for decoupling cognitive states, which comprises a problem and concept embedding representation module, a cognitive state decoupling module and a cognitive state tracking module; The problem and concept embedding representation module is used for constructing a problem-concept heterogeneous relation diagram and a triplet relation thereof, learning the embedding representation of the problem and concept according to the triplet relation, obtaining a problem vector and a concept vector, and obtaining basic interaction embedding according to interaction; The cognitive state decoupling module is used for constructing a fluctuation cognitive state and a stable cognitive state; the cognitive state tracking module is used for fusing the fluctuation cognitive state and the stable cognitive state and carrying out attenuation attention tracking to obtain prediction probability. The invention also provides a knowledge tracking method of the decoupling cognitive state, which utilizes the knowledge tracking system of the decoupling cognitive state to carry out knowledge tracking. The knowledge tracking method and system for decoupling cognitive states provided by the invention have the following beneficial effects: According to the invention, a learner cognitive state is decoupled into a fluctuation cognitive state and a stable cognitive state, two paths are respectively modeled, short-term cognitive fluctuation and long-term knowledge accumulation are prevented from being mixed together by a single hidden vector, explicit decoupling of the cognitive s