Enhanced Plagiarism Detection in Online Assignments by Natural Language Processing and Machine Learning Techniques

As online projects become more common in the educational setting, ensuring academic integrity is a growing concern. Traditional plagiarism detection algorithms often have trouble meeting some of the more sophisticated practices that students adopt for avoiding detection such as paraphrasing or word substitution. We present a machine learning based approach that was designed to enhance the effectiveness and accuracy of online assignment plagiarism detection, the proposed system can also detect minor plagiarism cases like paraphrased information and cross-linguistically similar. Due to the fact that the model proposed is capable of learning complex patterns and it was trained on a large dataset of a marginally homogenous dataset (academic publications), it maintains relatively high rates of recall and precision. The system uses multiple approaches of information retrieval to combine the interactions of textual relationships and elements such that it captures the text semantics of how person is feeling through word embedding. Later, it trains supervised learning models to classify possible plagiarism cases, like neural networks and Support Vector Machines (SVM). Also, the model can learn continuously from new data to improve its results. Our method improves the detection of traditional plagiarism and is also effective for more complex and subtle types of content reuse based on preliminary evaluation results.