Dynamic perception test question recommendation method and system fused with deep knowledge tracking

Abstract

The invention relates to a dynamic perception test question recommendation method and system fused with deep knowledge tracking, and belongs to the field of computers. The method comprises the following steps of: 1, inputting test question interaction sequence data of students into an embedding layer, converting large-scale sparse features into low-dimensional dense features suitable for model input, and interactively sequencing test questions of the students according to time; 2, modeling a dependency relationship between test question interaction sequences by using a gating loop unit, and introducing an auxiliary loss function to supervise learning in a current hidden state by utilizing next test question interaction; 3, designing a gating circulation unit with a self-attention updating gate to model a knowledge evolution process related to the target test question; and 4, connecting the dynamic knowledge state of the student and the embedded vector of the target test question in series, and feeding into a multi-layer perceptron to carry out final Softmax judgment to determine whether the target test question needs to be recommended to the student or not. According to the method, the performance of test question recommendation prediction is effectively improved, and personalized test question recommendation for students is realized.

Description

technical field [0001] The invention belongs to the field of computers, and relates to a method and system for recommending dynamic perception test questions integrated with deep knowledge tracking. Background technique [0002] With the development of education informatization, online learning represented by Massive Open Online Course platform (Massive Open Online Course, MOOC) and Intelligent Tutoring System (Intelligent Tutoring System, ITS), due to the convenience of two-way interaction and the The characteristics of richness and diversity of teaching interaction are more and more favored by people. However, due to the huge resources of test questions in the information system of online learning, it is impossible for students to practice all test questions in a limited time. Therefore, how to recommend suitable test questions for students among the massive test question resources is a key problem to be solved urgently in the intelligent information system in the field o...

AI Technical Summary

Problems solved by technology

[0005] However, the current research still has the following defects: (1) Most models directly use the interaction sequence of students' test questions to model the state of students' knowledge, and due to the influence of students' external environment (such as noise outside the classroom) and internal cognition (such as carelessness, mentality), As a result, the underlying knowledge state is difficult to be reflected by the explicit test item interaction sequence, so there is a lack of modeling of the underlying knowledge mastery behind the specific test item interaction sequence

This may lead to the phenomenon of "knowledge switching", that is, the knowledge points contained in the students' adjacent test questions may be very different, which makes students frequently switch from "function" knowledge points to "collection" knowledge in the process of doing the questions Click to switch to other knowledge points, different test papers form different answering sequences

Due to the influence of the "knowledge switching" phenomenon, students' answers to the target test questions may depend on knowledge points from a long time ago, adjacent knowledge points, or multiple knowledge points in different locations; The disadvantage of relative order effects, that is, when two identical test questions are answered in different orders, the prediction results of the model may be different

Images