Abnormal Profiles Detection Based on Time Series and Target Item Analysis for Recommender Systems

Abstract EN

Collaborative filtering (CF) recommenders arevulnerable to shilling attacks designed to affect predictions becauseof financial reasons.

Previous work related to robustnessof recommender systems has focused on detecting profiles.

Mostapproaches focus on profile classification but ignore the groupattributes among shilling attack profiles.

Attack profiles areinjected in a short period in order to push or nuke a specifictarget item.

In this paper, we propose a method for detectingsuspicious ratings by constructing a time series.

We reorganizeall ratings on each item sorted by time series.

Each timeseries is examined and suspected rating segments are checked.

Then we use techniques we have studied in previous studyto detect shilling attacks in these anomaly rating segmentsusing statistical metrics and target item analysis.

We show inexperiments that our proposed method can be effective and lesstime consuming at detecting items under attacks in big datasets.