Title: Alternative Adversarial Sampling Attack on Text and Sentiment Classification
• Study and evaluate existing representations and classifications on text and document.
• Study adversary model on adversarial method towards machine learning algorithm.
• Build and evaluate similar existing attack method.
• Simulate practical and realistic scene of adversarial attack on classification model.
• Build Machine Learning model on NLP classification task.
• Develop new method of word transformer for word-level attack.
• Optimize influential word sorting by precedency score.
• Achieve best results of attack efficacy on Bi-directional LSTM and BERT.
• Achieve best results in terms of perturbation rate.