HHFHNet: Hybrid Deep Learning Network for Course Recommendation Using H-Matrix

Abstract

Background

Students often need help choosing the right courses to complete their degrees. Course recommender systems assist in selecting suitable academic courses. Recent attention-based have been developed to distinguish the influence of past courses on recommendations. However, these models might not work well when users have diverse interests, because the effectiveness of the attention mechanism decreases with the variety of historical courses. To overcome these issues, this study introduces a new approach called Hierarchical Attention Network with Hierarchical Deep Learning for Text Forward Harmonic Net (HHFHNet) for course recommendations using H-matrix.

Methods

Initially, the input course data obtained from the dataset is processed into course overview and course genres. After that, the Term Frequency-Inverse Document Frequency (TF-IDF) method is applied to both the course overview and query, with the resulting output fed into the HHFHNet, which combines Hierarchical Deep Learning for Texts (HDLTex) and Hierarchical Attention Networks (HAN). This generates a Course Recommendation Probability Value (CRPV), which is used to retrieve recommended courses. Simultaneously, specific course genre features are selected using chord distance. Then, specific course genre features are selected using chord distance. These selected features and CRPV are then used with the H-matrix to create ranking-based recommendations. Finally, Explainable Artificial Intelligence (XAI) is utilized to generate course recommendation messages based on the ranking approach.

Results

The effectiveness of the HHFHNet technique was evaluated using performance metrics such as precision, recall, and F-measure, and it achieved values of 90.31%, 91.87%, and 91.08%, respectively.

Conclusions

The proposed HHFHNet technique significantly enhances course recommendation accuracy and offers a robust solution for guiding students in their academic course selection.