Automated essay scoring with SBERT embeddings and LSTM-Attention networks

Area of Science:

• Educational Technology
• Natural Language Processing
• Artificial Intelligence in Education

Background:

• Automated essay scoring (AES) is crucial for efficient and objective student writing evaluation.
• Existing AES methods face challenges in capturing nuanced linguistic features and contextual meaning.
• There is a need for more sophisticated models to improve the accuracy and reliability of automated assessments.

Purpose of the Study:

• To develop and evaluate an innovative AES method.
• To enhance scoring accuracy by integrating advanced deep learning techniques.
• To improve the reliability and efficiency of automated writing evaluation systems.

Main Methods:

• Integration of Sentence-BERT (SBERT) for generating essay embeddings.
• Utilization of bidirectional Long Short-Term Memory (BiLSTM) networks to process sequential embedding data.
• Incorporation of an attention mechanism to focus on salient essay components.

Main Results:

• The proposed SBERT-BiLSTM with attention model demonstrated significant improvements in scoring accuracy.
• The system effectively learned and utilized features from essay embedding vectors.
• The attention mechanism enhanced the model's ability to prioritize critical textual elements.

Conclusions:

• The developed AES approach offers a substantial advancement in automated writing assessment.
• This method provides a more reliable and efficient alternative to traditional scoring techniques.
• The findings highlight the potential of combining SBERT, BiLSTM, and attention for educational technology applications.