Privacy-Preserving Sequential Recommendation with Collaborative Confusion

TL;DR

CLOUD introduces a privacy-preserving sequential recommender by collaboratively confusing a user’s interaction sequence through neighboring sequences before recommendation. It combines a shared encoder, an item-wise modifier with a copy mechanism, and a BERT4Rec–style recommender, training with a joint SSL objective on both raw and modified sequences; the system achieves a high modification rate (up to $66\%$) while maintaining competitive accuracy across real-world datasets. Empirical results show CLOUD delivers state-of-the-art or near-state-of-the-art recommendation performance with significantly reduced privacy leakage, and it remains robust under simulated data destruction and across varying recommender back-ends. The work suggests CLOUD as a practical, communication-efficient privacy-preserving approach for sequential recommendations, with future directions including additional noise mechanisms and adversarial training.

Abstract

Sequential recommendation has attracted a lot of attention from both academia and industry, however the privacy risks associated to gathering and transferring users' personal interaction data are often underestimated or ignored. Existing privacy-preserving studies are mainly applied to traditional collaborative filtering or matrix factorization rather than sequential recommendation. Moreover, these studies are mostly based on differential privacy or federated learning, which often leads to significant performance degradation, or has high requirements for communication. In this work, we address privacy-preserving from a different perspective. Unlike existing research, we capture collaborative signals of neighbor interaction sequences and directly inject indistinguishable items into the target sequence before the recommendation process begins, thereby increasing the perplexity of the target sequence. Even if the target interaction sequence is obtained by attackers, it is difficult to discern which ones are the actual user interaction records. To achieve this goal, we propose a CoLlaborative-cOnfusion seqUential recommenDer, namely CLOUD, which incorporates a collaborative confusion mechanism to edit the raw interaction sequences before conducting recommendation. Specifically, CLOUD first calculates the similarity between the target interaction sequence and other neighbor sequences to find similar sequences. Then, CLOUD considers the shared representation of the target sequence and similar sequences to determine the operation to be performed: keep, delete, or insert. We design a copy mechanism to make items from similar sequences have a higher probability to be inserted into the target sequence. Finally, the modified sequence is used to train the recommender and predict the next item.

Figures (6)

• Figure 1: Schematic diagram of the recommendation process of CLOUD. The recommender only stores the modified interaction sequence, and even if it is attacked, the real interaction sequence will not be leaked.
• Figure 2: Overview of CLOUD. The encoder is used to encode the target raw sequence and similar sequence into hidden representation. When training the item-wise modifier, it is required to perform delete operation and insert operation to restore the randomly modified sequence. The target sequence and its modified version are used together to train the recommender. Finally, CLOUD jointly optimizes the loss functions of the item-wise modifier and recommender.
• Figure 3: Self-modifier of STEAM. When inserting, the self-modifier uses a reverse generator to insert several items in reverse in front of the target position.
• Figure 4: Schematic diagram of the probability distribution of three modification operations calculated by CLOUD.
• Figure 5: The copy mechanism of item-wise modifier. Items from similar sequences have a higher probability of being inserted overall.