Posted by Admin: System Admin
In e-commerce, online retailers are usually suffering from professional malicious users (PMUs), who utilize negative reviews and low ratings to their consumed products on purpose to threaten the retailers for illegal profits. PMUs are difficult to be detected because they utilize masking strategies to disguise themselves as normal users. Specifically, there are three challenges for PMU detection: 1) professional malicious users do not conduct any abnormal or illegal interactions (they never concurrently leave too many negative reviews and low ratings at the same time), and they conduct masking strategies to disguise themselves. Therefore, conventional outlier detection methods are confused by their masking strategies. 2) the PMU detection model should take both ratings and reviews into consideration, which makes PMU detection a multi-modal problem. 3) there are no datasets with labels for professional malicious users in public, which makes PMU detection an unsupervised learning problem. To this end, we propose an unsupervised multi-modal learning model: MMD, which employs Metric learning for professional Malicious users Detection with both ratings and reviews. MMD first utilizes a modified RNN to project the informational review into a sentiment score, which jointly considers the ratings and reviews. Then professional malicious user profiling (MUP) is proposed to catch the sentiment gap between sentiment scores and ratings. MUP filters the users and builds a candidate PMU set. We apply a metric learning-based clustering to learn a proper metric matrix for PMU detection. Finally, we can utilize this metric and labeled users to detect PMUs. Specifically, we apply the attention mechanism in metric learning to improve the model’s performance. The extensive experiments in four datasets demonstrate that our proposed method can solve this unsupervised detection problem. Moreover, the performance of the state-of-the-art recommender models is enhanced by taking MMD as a preprocessing stage. Machine learning is an important component of the growing field of data science. Through the use of statistical methods, different type of algorithms is trained to make classifications or predictions, and to uncover key insights in this project. These insights subsequently drive decision making within applications and businesses, ideally impacting key growth metrics. Machine learning algorithms build a model based on this project data, known as training data, in order to make predictions or decisions without being explicitly programmed to do so. Machine learning algorithms are used in a wide variety of datasets, where it is difficult or unfeasible to develop conventional algorithms to perform the needed tasks.
As we define professional malicious users in recommender systems, malicious user detection is a new problem, which is an issue with little attention yet. However, we can treat this detection issue as a special case of abnormal user detection, and some existing works in this area can inspire us [35, 36]. In e-commerce, various abnormal users (spammers, shilling group, and frauds) have greatly damaged the systems, and some abnormal user detection models are proposed to tackle this issue. [37] proposed a hybrid model to detect the spammers through users’ profile and relations. [38] explored spammer detection in big and sparse data. Shilling attacks harm the recommender system by injectingfake profile information of users and items. They cheat the recommendation model, such as Collaborative Filtering and Matrix Factorization [12]. [39] proposed this attack type and gave a basic supervised solution to tackle it. [12] proposed a convolutional neural network to solve shilling attacks and improved collaborative filtering. Frauds usually give fake reviews to hurt the profits of electronic retailers. [40, 41] also explored fraud detection in large-scale dataset and real scenarios. Some researches of abnormal user detection utilize the machine learning model to find fake ratings or reviews [42, 43] and achieve an effective result. However, different from abnormal users above (spammers, shilling group, and fraud), professional malicious users are smarter and craftier. Shilling attacks inject fake ratings or reviews just before the recommendation process [44, 45], while for PMUs, all the actions that professional malicious users have taken are well-behaved by the rules of e-commerce websites (called masking strategies). They utilize the bug of abnormal detections, without leaving low ratings and negative feedback at the same time, to avoid detections. Then they can make illegal profits and hurt the electronic retailers. Basically, they are “normal” users for the existing abnormal user detection models, which makes the professional malicious user detection a critical issue in the recommender system area. Disadvantages ? The system not implemented Professional Malicious User Profiling (MUP) model. ? The system not implemented Attention Metric Learning for Clustering (MLC) and Hierarchical Dual-Attention RNN.
_ This is the first work focusing on solving the professional malicious user detection issue utilizing both users’ ratings and reviews to enhance the state-ofthe- art recommender systems. _ A novel multi-modal unsupervised method-MMD-is proposed to detect professional malicious users with the modified RNN and attention metric learning based clustering. _ Extensive experiments are conducted on four real world e-commerce datasets to verify our proposed method. Moreover, by filtering professional malicious users, some state-of-the-art models are enhanced. Advantages _ This is the first work focusing on solving the professional malicious user detection issue utilizing both users’ ratings and reviews to enhance the state-ofthe- art recommender systems. _ A novel multi-modal unsupervised method-MMD-is proposed to detect professional malicious users with the modified RNN and attention metric learning based clustering. _ Extensive experiments are conducted on four real world e-commerce datasets to verify our proposed method. Moreover, by filtering professional malicious users, some state-of-the-art models are enhanced.