A Bidirectional LSTM Language Model for Code Evaluation and Repair

Programming is a vital skill in computer science and engineering-related disciplines. However, developing source code an error-prone task. Logical errors are particularly hard to identify for both students professionals, single error unexpected end-users. At present, conventional compilers have difficulty identifying many of the (especially logical errors) that can occur code. To mitigate this problem, we propose language model evaluating codes using bidirectional long short-term memory (BiLSTM) neural network. We trained BiLSTM with large number tuning various hyperparameters. then used evaluate incorrect assessed model’s performance three principal areas: detection, suggestions repair, erroneous classification. Experimental results showed proposed achieved 50.88% correctness providing suggestions. Moreover, F-score approximately 97%, outperforming other state-of-the-art models (recurrent networks (RNNs) (LSTM)).