Deep learning knowledge tracking method and system based on attention mechanism

Abstract

The invention belongs to the field of knowledge tracking, and provides a deep learning knowledge tracking method and system based on an attention mechanism, which can reserve context sequence information by adopting a self-attention mechanism, explicitly call question interaction in the whole historical sequence of students, and improve the knowledge tracking efficiency. Different weights among elements in long and short-term historical sequences are clearly calculated, dynamic long-term and short-term knowledge states are captured, the long-term and short-term knowledge states of students are accurately captured and enhanced in the form of subsequence division, and short-term knowledge representation of the students is reflected by continuous behaviors of the students in the current time span. All historical behaviors before the current time span reflect the long-term knowledge state of the student, and the self-attention mechanism can adaptively learn the corresponding dependency relationship of the long-term and short-term continuous behaviors of the student.

Description

technical field [0001] The invention belongs to the field of knowledge tracking, and in particular relates to a deep learning knowledge tracking method and system based on an attention mechanism. Background technique [0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art. [0003] The purpose of knowledge tracking is to dynamically model the personalized learning process of students' learning and answering sequences in the form of models, and to mine the relationship between knowledge points and students' dynamic grasp of knowledge points. In the past ten years, many knowledge tracking models have been proposed, most of which are based on recurrent neural networks. [0004] With the successful modeling of deep learning in the fields of natural language processing and sequence recommendation, many researchers have applied it to the field of knowledge tracking, which is called ...

AI Technical Summary

Problems solved by technology

[0006] Long Short-Term Memory (LSTM) networks rely on the student's behavior at the previous timestep when processing sequences, and historical influences are quickly lost for several timesteps, thus failing to capture long-term and learned performance

Recurrent neural network (RNN) based methods tend to compress the student's history into a fixed hidden state vector, which cannot explicitly capture the item interaction regardless of the distance between two items in the sequence

[0007] The cyclic neural network is more dependent on the previous time step modeling than the recent time segment modeling, and cannot model the knowledge status of students in the short-term time segment.

[0008] In summary, existing approaches to knowledge tracking model dependencies without considering sequence positions, leading to loss of temporal information

As such, they are insufficient to describe and distinguish students' dynamic and changing long-term and short-term learning abilities, and cannot accurately predict students' knowledge acquisition

Images