Design and Application of Digital Resource Library for Japanese Language Teaching based on Cloud Technology using BERT-NCF

Recent days, the increasing demand for Japanese language instruction has created a need for comprehensive digital resources to support teaching and learning. A digital library provides accessible and engaging materials for educators and students. Traditional methods for building a digital resource library for Japanese language has faced significant challenges like the data contains irrelevant features and quality of text is poor. Therefore, this paper proposes Bidirectional Encoder Representations from Transformers - Neural Collaborative Filtering (BERT-NCF) for building a digital resource library for Japanese language teaching which is employed on Jisho API dataset which consists of words, phrases and sentences of Japanese language. Then, the data is preprocessed by using Natural Language Toolkit (NLTK) approaches like tokenization, stop word removal, stemming and Lemmatization which improved the text quality and reduced noise. After that, the optimal features are extracted by using BERT which effectively eliminated the irrelevant features from data and model accuracy is improved by using Adam Weight decay (AdamW) optimizer. Finally, NCF is used to recommend useful personalized Japanese language learning resources by using obtained features. The proposed BERT-NCF achieved better results in terms of accuracy ($\mathbf{9 8. 5 0 \%}$), Precision (91.25%), Recall (92.49%), and F1 score (95.26%) when compared with existing BERT model.