HMSTNet: A Deep Learning Multimodal Approach for Personalized English Literature Recommendations

Abstract

Recommendation systems serve as fundamental elements of contemporary technology by reshaping how users find content in all fields including educational applications. Academic English literature benefits strongly from recommendation systems because these systems enable customized learning opportunities to help students discover relevant content according to personal learning needs. English literature represents features challenging recommendation systems because it consists of extensive text content combined with multiple stylistic attributes and extensive metadata structures. The study presents a deep learning framework called HMSTNet to manage complex recommendation scenarios. Novel architecture integrates metadata datasets alongside stylistic information and BERT-powered text-based embeddings for understanding of content and user systemic preferences. Through its multi-branch structure Hybrid Multimodal Semantic Text Neural Network (HMSTNet) based on LSTM model to recognize difficult meta-relations between text components while utilizing user embeddings to maintain context awareness. This research serves to unite conventional recommendation algorithms with English literature reading requirements. The complex nature of literary data poses challenges to k-Nearest Neighbors (kNN) and Singular Value Decomposition (SVD) whereas conventional methods show limited success in both context understanding and accuracy. The model HMSTNet outperforms others with outstanding results reaching 95.23% accuracy. Multiple metrics demonstrate the efficiency of this model because they measure highest MBD with KNN model of 6.8m along with Theil’s U-statistic of 0.04 and 90th percentile error is 8.41 to confirm minimized prediction errors and enhanced accuracy and fail-safe capability. Through its reliable framework HMSTNet both delivers personalized content recommendations and supports educational technology development which deepens student engagement with literature and promotes lifelong learning.

Keywords

• BERT embeddings
• deep-learning
• English literature
• metadata
• recommendation system
• users personalization