Optimizing Automated Scoring for Hindi Responses: A Hybrid Framework with Advanced Feature Integration

The educational assessment field undergoes a transformation through automated answer scoring (AAS) technology which allows efficient real-time student response evaluation. Current approaches for linguistic and contextual detection in Hindi tend to fall short because this language shows complex morphology and syntactical diversity. This restriction creates problems with accuracy and reliability in scoring systems when dealing with non-English datasets at large. We resolve linguistic issues through semantic and syntactic elements that incorporate word embeddings and have developed an RNN-LSTM and XGBoost hybrid model. The proposed methodology employs RNN- long short-term memory (LSTM) because it recognizes contextual relationships in order to understand word sequences, while XGBoost provides robust ensemble learning to model nonlinear data relationships. A dataset composed of Hindi questions and their reference and student answers undergoes [Formula: see text]-fold cross-validation for model verification. A hybrid model for Automated Automatic Scoring of AAS in Hindi proved effective by using both syntactic and semantic elements with rules and artificial intelligence (AI) approaches to develop a flexible scoring system that matches human evaluation standards while exhibiting strong precision, recall, accuracy, and F1 values along with high correlation and substantial agreement.