Cognitive-Based Hybrid Collaborative Filtering with Rating Scaling on Entropy to Defend Shilling Influence.

In the current era of big data, huge volumes a wide variety of valuable data are generated and collected at a high velocity. Hence, data science solutions are in demand to data mine these big data for valuable information and useful knowledge embedded in these big data in order to transform this information and knowledge into recommendations and actions. In particular, recommendation systems (RecSys or RS)---which are tools that can provide suggestions to users based on various metrics---have been playing an important role in society since the booming of the Internet. Making more accurate predictions can both potentially increase company revenue and enhance user experience. So, it has been a hot topic. More specifically, collaborative filtering (CF) has been a popular technique applied in RS. The key ideas behind most of the CF algorithms are to filter items based on other users' opinions. Since the recommendation process is based on user interactions, one of the challenges is how to prevent shilling attacks (or shilling attack ratings). In this paper, we propose methods to integrate users' rating entropy into collaborative filtering so as to defend shilling attacks and reduce noisy ratings, and thus achieve higher prediction accuracy. Evaluation results show the effectiveness of our cognitive-based hybrid collaborative filtering methods in rating scaling on entropy for defending shilling influence.