FABRIC - A Fault-aware BiLSTM Approach for Programming Bug Detection and Corrective Suggestion

This paper addresses the crucial challenge of automated detection and correction of programming logic errors for students. While syntax errors are handled by modern IDEs, logic errors remain a significant hurdle. Existing Automated Program Repair (APR) techniques in educational settings have limitations due to wasting large amounts of incorrect source code. To overcome this, we introduce FABRIC, a fault-aware BiLSTM approach that enhances error detection by leveraging buggy-fixed code pairs and integrating score information from user submissions. FABRIC identifies error locations and suggests fixes using both correct and erroneous code. Evaluated on datasets from the AIZU Online Judge (AOJ) and the Programming Fundamentals Course (PFC) at Ho Chi Minh City University of Technology, FABRIC demonstrates reduced sensitivity to noisy or incorrect code compared to the original BiLSTM model, leading to more stable and reliable error detection. This improvement is reflected in enhanced global accuracy metrics, confirming FABRIC’s effectiveness in consistently identifying and addressing logic errors across diverse programming submissions.