Answers are grounded in this paper and cite their sources.
Behavioral fingerprints refer to the unique patterns and characteristics of user behaviors hidden in a number of activity traces generated in the digital space. It is difficult to forge and supports non-intrusive data collection, making it well-suited for user identification. However, the sparsity of behavioral data often leads to unsatisfactory identification performance. This paper adopts the frequency of user access to content items as the behavioral fingerprint and proposes two data augmentation methods to address this issue. On one hand, we update the unsupervised behavioral fingerprinting method from SURE to SURE+, which utilizes the union of n-level extendable multi-item-set fingerprints to preserve more user features. On the other hand, we have designed a convolutional neural network that is trained on mixup data, surpassing the performance of models trained solely on original data. Additionally, we have constructed a fusion decision-making model based on the outputs of unsupervised methods and supervised learning approaches to further enhance identification performance. Experimental results from three distinct datasets demonstrate that our proposed methods significantly outperform existing approaches in terms of identification accuracy. Especially, we achieved close to 100% accuracy on small-scale datasets with high frequency access.
No takes yet. Share an insight, caveat, or question.
Yang et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: