Blackmailer or Consumer? A Character-level CNN Approach for Identifying Malicious Complaint Behaviors

Blackmailers are becoming a serious Ecommerce security problem in China. They use the threat of complaining in an extortion ploy which is disrupting normal business order and causing interference in law enforcement and justice. This paper proposed a character-level CNN approach, named DeepDetector, to identify malicious complaint behaviors. DeepDetector was a simple CNN with one layer of convolution on top of character vectors obtained from lookup table. The experimental results showed that the AUC and F1 score of DeepDetector were respectively 97.32% and 96.00%. Both higher than that of RNN and bag-of-words, bag-of-n-grams model using machine learning method like Random Forest or XGBoost as classifier. Based on the real-world dataset of “National 12358 price regulation platform of China”, the data-driven experimental results indicated the effectiveness and efficiency of DeepDetector. Finally, the model was implemented in the whole unlabeled dataset ranging from Apr 1st, 2015 to Apr 1st, 2019 to identify malicious complaint behaviors. We analyzed the behavioral patterns and gathering area of E-commerce blackmailers in detail. Our study is useful for market regulators to identify E-commerce blackmailers and allocate supervision resources.