Construction and Application of Personalized Teaching Algorithm Model Based on Time Series Data Analysis

Abstract

In online education scenarios, student learning behavior data exhibits significant temporal characteristics, but existing teaching systems struggle to capture the dynamic changes in students' knowledge mastery, leading to a lack of targeted teaching resource recommendations. This paper constructs a temporal analysis model based on the fusion of Long Short-Term Memory (LSTM) networks and attention mechanisms to achieve personalized teaching path planning. First, 12-dimensional behavioral features, including student answer accuracy, dwell time, and number of repeated learning sessions, are collected to construct a 30-day sequence dataset. Second, a three-layer LSTM network is designed to extract temporal dependencies, and the weight distribution at different time steps is calculated through a self-attention layer to identify key learning nodes. Then, knowledge graph mapping technology is combined to correlate students' ability status with their knowledge mastery in real time. Finally, differentiated learning plans are generated based on the prediction results. Validated on a math subject dataset covering 8,500 middle school students, the model reduced its prediction error for knowledge point mastery to 0.127 (root mean square error). The average learning time per knowledge point for students in the experimental group is reduced from 104.4 minutes to 70.8 minutes. Targeted training on weak knowledge points improved the score of the severely deficient group from 48.3 to 68.1, an improvement rate of 41.0%. The experiment shows that this method can effectively characterize the dynamic evolution of the learning process and provide accurate decision support for intelligent education systems.