AI-Based Educational Interactive Agents Using Hybrid Query Classification

One of the most important aspects of today’s educational landscape is the role that digital platforms play in enhancing students’ educational experiences. Integrating AI is crucial for correctly interpreting and categorizing student inquiries. This study introduces query-based classification using advanced neural networks. In order to do this, data are initially collected from the Stanford Question Answering dataset, which is a compilation of student inquiries sourced from educational platforms and social media. After obtaining this dataset, the data pre-processing will be done by using text case normalization, tokenization, stop word removal, and lemmatization to distill texts to their core components. After pre-processing, different types of word embedding models, like an improved TF-IDF model, Global Vectors for Word Representation (Glove)-V model, position encoding Continuous Bag-of-Words (PE-CBOW) model, and an enhanced Word2vec used to represent the words into a fixed vector representation. Finally, the student’s queries were categorized using the proposed novel Hybrid Spatial Attention-assisted Long Short-Term Memory (LSTM)-enabled Bidirectional Encoder Representations from Transformers (BERT) (HA-LSTM-BERT) model. The proposed classification models will effectively categorize students’ input queries by identifying valuable phrases for their queries. The investigational results of the proposed HA-LSTM-BERT model attain an accuracy of 98.7%, which is effective in query classification.