DP-MFRNN: Difficulty Prediction for Examination Questions Based on Neural Network Framework

The conventional approach to predicting examination question difficulty primarily relies on artificial marking, characterized by high subjectivity, low accuracy, and substantial workload. This approach significantly hampers the progress and advancement of intelligent education evaluation systems. In order to address these challenges, we propose DP-MFRNN, a bidirectional recurrent neural network model based on multi-feature attention. The DP-MFRNN model adopts a multi-feature task learning strategy, leveraging relevant computational knowledge to enhance the contextual understanding of the problem statement. Subsequently, a bidirectional recurrent neural network is employed to mine the logical relations within the test text data and extract the representation of the statement. The attention mechanism is then utilized to measure the importance of associated statements to the given problem. Taking the example of a university computer foundation single-item selection, the extracted features are input into the model. Experimental results demonstrate a significant enhancement in the Pearson correlation coefficient and the degree of agreement of the proposed model.