Recommender system plays a pervasive role in today’s online services, yet its closed-loop nature, i.e., training and deploying within a specific closed domain, constrains its access to open-world knowledge. Recently, the emergence of large language models (LLMs) has shown promise in bridging this gap by encoding extensive world knowledge and demonstrating advanced reasoning capabilities. However, previous attempts to directly implement LLMs as recommenders fall short in meeting the demanding requirements of industrial recommender systems, particularly in terms of online inference latency and offline resource efficiency. In this work, we propose an Open-World R ecommendation Framework with E fficient and Deployable K nowledge I nfusion from Large Language Models, dubbed REKI , to acquire two types of external knowledge about users and items from LLMs. Specifically, we introduce factorization prompting to elicit accurate knowledge reasoning on user preferences and items. With factorization prompting, we develop individual knowledge extraction and collective knowledge extraction tailored for different scales of recommendation scenarios, effectively reducing offline resource consumption. Subsequently, the generated user and item knowledge undergoes efficient transformation and condensation into augmented vectors through a hybridized expert-integrated network , ensuring its compatibility with the recommendation task. The obtained vectors can then be directly used to enhance the performance of any conventional recommendation model. We also ensure efficient inference by preprocessing and prestoring the knowledge from the LLM. Extensive experiments demonstrate that REKI significantly outperforms the state-of-the-art baselines and is compatible with a diverse array of recommendation algorithms and tasks. Now, REKI has been deployed to Huawei’s news and music recommendation platforms and gained a 7% and 1.99% improvement during the online A/B test.

Accurately predicting the probabilities of user feedback, such as clicks and conversions, is critical for advertisement ranking and bidding. However, there often exist unwanted mismatches between predicted probabilities and true likelihoods due to the rapid shift of data distributions and intrinsic model biases. Calibration aims to address this issue by post-processing model predictions, and field-aware calibration can adjust model output on different feature field values to satisfy fine-grained advertising demands. Unfortunately, the observed samples corresponding to certain field values can be seriously limited to make confident calibrations, which may yield bias amplification and online disturbance. In this paper, we propose a confidence-aware multi-field calibration method, which adaptively adjusts the calibration intensity based on confidence levels derived from sample statistics. It also utilizes multiple fields for joint model calibration according to their importance to mitigate the impact of data sparsity on a single field. Extensive offline and online experiments show the superiority of our method in boosting advertising performance and reducing prediction deviations.

Recommender system plays a vital role in various online services. However, its insulated nature of training and deploying separately within a specific closed domain limits its access to open-world knowledge. Recently, the emergence of large language models (LLMs) has shown promise in bridging this gap by encoding extensive world knowledge and demonstrating reasoning capabilities. Nevertheless, previous attempts to directly use LLMs as recommenders cannot meet the inference latency demand of industrial recommender systems. In this work, we propose an Open-World Knowledge Augmented Recommendation Framework with Large Language Models, dubbed KAR, to acquire two types of external knowledge from LLMs — the reasoning knowledge on user preferences and the factual knowledge on items. We introduce factorization prompting to elicit accurate reasoning on user preferences. The generated reasoning and factual knowledge are effectively transformed and condensed into augmented vectors by a hybrid-expert adaptor in order to be compatible with the recommendation task. The obtained vectors can then be directly used to enhance the performance of any recommendation model. We also ensure efficient inference by preprocessing and prestoring the knowledge from the LLM. Extensive experiments show that KAR significantly outperforms the state-of-the-art baselines and is compatible with a wide range of recommendation algorithms. We deploy KAR to Huawei’s news and music recommendation platforms and gain a 7% and 1.7% improvement in the online A/B test, respectively.

Tabular data is one of the most common data storage formats behind many real-world web applications such as retail, banking, and e-commerce. The success of these web applications largely depends on the ability of the employed machine learning model to accurately distinguish influential features from all the predetermined features in tabular data. Intuitively, in practical business scenarios, different instances should correspond to different sets of influential features, and the set of influential features of the same instance may vary in different scenarios. However, most existing methods focus on global feature selection assuming that all instances have the same set of influential features, and few methods considering instance-wise feature selection ignore the variability of influential features in different scenarios. In this paper, we first introduce a new perspective based on the influence function for instance-wise feature selection, and give some corresponding theoretical insights, the core of which is to use the influence function as an indicator to measure the importance of an instance-wise feature. We then propose a new solution for discovering instance-wise influential features in tabular data (DIWIFT), where a self-attention network is used as a feature selection model and the value of the corresponding influence function is used as an optimization objective to guide the model. Benefiting from the advantage of the influence function, i.e., its computation does not depend on a specific architecture and can also take into account the data distribution in different scenarios, our DIWIFT has better flexibility and robustness. Finally, we conduct extensive experiments on both synthetic and real-world datasets to validate the effectiveness of our DIWIFT.

How to effectively mitigate the bias of feedback in recommender systems is an important research topic. In this article, we first describe the generation process of the biased and unbiased feedback in recommender systems via two respective causal diagrams, where the difference between them can be regarded as the source of system-induced biases. We then define this difference as a confounding bias and propose a new perspective on debiased representation learning to alleviate it. Specifically, for the case with biased feedback alone, we derive the conditions that need to be satisfied to obtain a debiased representation from the causal diagrams. Then, we propose a novel framework called debiased information bottleneck (DIB) to optimize these conditions and then find a tractable solution for it. The proposed framework constrains the model to learn a biased embedding vector with independent biased and unbiased components in the training phase, and uses only the unbiased component in the test phase to deliver more accurate recommendations. We further propose a variant of DIB by relaxing the independence between the biased and unbiased components. Finally, we conduct extensive experiments on a public dataset and a real product dataset to verify the effectiveness of the proposed framework.

Due to the poor generalization performance of traditional empirical risk minimization (ERM) in the case of distributional shift, Out-of-Distribution (OoD) generalization algorithms receive increasing attention. However, OoD generalization algorithms overlook the great variance in the quality of training data, which significantly compromises the accuracy of these methods. In this paper, we theoretically reveal the relationship between training data quality and algorithm performance, and analyze the optimal regularization scheme for Lipschitz regularized invariant risk minimization. A novel algorithm is proposed based on the theoretical results to alleviate the influence of low quality data at both the sample level and the domain level. The experiments on both the regression and classification benchmarks validate the effectiveness of our method with statistical significance.

Click-through rate (CTR) prediction model usually consists of three components: embedding table, feature interaction layer, and classifier. Learning embedding table plays a fundamental role in CTR prediction from the view of the model performance and memory usage. The embedding table is a two-dimensional tensor, with its axes indicating the number of feature values and the embedding dimension, respectively. To learn an efficient and effective embedding table, recent works either assign various embedding dimensions for feature fields and reduce the number of embeddings respectively or mask the embedding table parameters. However, all these existing works cannot get an optimal embedding table. On the one hand, various embedding dimensions still require a large amount of memory due to the vast number of features in the dataset. On the other hand, decreasing the number of embeddings usually suffers from performance degradation, which is intolerable in CTR prediction. Finally, pruning embedding parameters will lead to a sparse embedding table, which is hard to be deployed. To this end, we propose an optimal embedding table learning framework OptEmbed, which provides a practical and general method to find an optimal embedding table for various base CTR models. Specifically, we propose pruning the redundant embeddings regarding corresponding features' importance by learnable pruning thresholds. Furthermore, we consider assigning various embedding dimensions as one single candidate architecture. To efficiently search the optimal embedding dimensions, we design a uniform embedding dimension sampling scheme to equally train all candidate architectures, meaning architecture-related parameters and learnable thresholds are trained simultaneously in one supernet. We then propose an evolution search method based on the supernet to find the optimal embedding dimensions for each field. Experiments on public datasets show that OptEmbed can learn a compact embedding table which can further improve the model performance.

Deep models often fail to generalize well in test domains when the data distribution differs from that in the training domain. Among numerous approaches to address this Out-of-Distribution (OOD) generalization problem, there has been a growing surge of interest in exploiting Adversarial Training (AT) to improve OOD performance. Recent works have revealed that the robust model obtained by conducting sample-wise AT also retains transferability to biased test domains. In this paper, we empirically show that sample-wise AT has limited improvement on OOD performance. Specifically, we find that AT can only maintain performance at smaller scales of perturbation while Universal AT (UAT) is more robust to larger-scale perturbations. This provides us with clues that adversarial perturbations with universal (low dimensional) structures can enhance the robustness against large data distribution shifts that are common in OOD scenarios. Inspired by this, we propose two AT variants with low-rank structures to train OOD-robust models. Extensive experiments on DomainBed benchmark show that our proposed approaches outperform Empirical Risk Minimization (ERM) and sample-wise AT. Our code is available at https://github.com/NOVAglow646/NIPS22-MAT-and-LDAT-for-OOD.

Sequential recommender systems aim to model users’ evolving interests from their historical behaviors, and hence make customized time-relevant recommendations. Compared with traditional models, deep learning approaches such as CNN and RNN have achieved remarkable advancements in recommendation tasks. Recently, the BERT framework also emerges as a promising method, benefited from its self-attention mechanism in processing sequential data. However, one limitation of the original BERT framework is that it only considers one input source of the natural language tokens. It is still an open question to leverage various types of information under the BERT framework. Nonetheless, it is intuitively appealing to utilize other side information, such as item category or tag, for more comprehensive depictions and better recommendations. In our pilot experiments, we found naive approaches, which directly fuse types of side information into the item embeddings, usually bring very little or even negative effects. Therefore, in this paper, we propose the NOn-inVasive self-Attention mechanism (NOVA) to leverage side information effectively under the BERT framework. NOVA makes use of side information to generate better attention distribution, rather than directly altering the item embeddings, which may cause information overwhelming. We validate the NOVA-BERT model on both public and commercial datasets, and our method can stably outperform the state-of-the-art models with negligible computational overheads.

How to effectively mitigate the bias of feedback in recommender systems is an important research topic. In this paper, we first describe the generation process of the biased and unbiased feedback in recommender systems via two respective causal diagrams, where the difference between them can be regarded as the source of bias. We then define this difference as a confounding bias, which can be regarded as a collection of some specific biases that have previously been studied. For the case with biased feedback alone, we derive the conditions that need to be satisfied to obtain a debiased representation from the causal diagrams. Based on information theory, we propose a novel method called debiased information bottleneck (DIB) to optimize these conditions and then find a tractable solution for it. In particular, the proposed method constrains the model to learn a biased embedding vector with independent biased and unbiased components in the training phase, and uses only the unbiased component in the test phase to deliver more accurate recommendations. Finally, we conduct extensive experiments on a public dataset and a real product dataset to verify the effectiveness of the proposed method and discuss its properties.

In the time of Big Data, training complex models on large-scale data sets is challenging, making it appealing to reduce data volume for saving computation resources by subsampling. Most previous works in subsampling are weighted methods designed to help the performance of subset-model approach the full-set-model, hence the weighted methods have no chance to acquire a subset-model that is better than the full-set-model. However, we question that how can we achieve better model with less data? In this work, we propose a novel Unweighted Influence Data Subsampling (UIDS) method, and prove that the subset-model acquired through our method can outperform the full-set-model. Besides, we show that overly confident on a given test set for sampling is common in Influence-based subsampling methods, which can eventually cause our subset-model's failure in out-of-sample test. To mitigate it, we develop a probabilistic sampling scheme to control the worst-case risk over all distributions close to the empirical distribution. The experiment results demonstrate our methods superiority over existed subsampling methods in diverse tasks, such as text classification, image classification, click-through prediction, etc.

Most commercial industrial recommender systems have built their closed feedback loops. Though it is helpful in item recommendation and model training, the closed feedback loop may lead to the so-called bias problems, including the position bias, selection bias and popularity bias. The recommendation models trained with biased may hurt the user experiences by recommending homogenous items. How to control the biases in the closed feedback loop has become one of major challenges in modern recommender systems. This talk discusses the counterfactual learning technologies for tackling the bias problem in recommendation.

Recommender system is one of the most successful machine learning technologies for commerce. However, it can reinforce the closed feedback loop problem, where the recommender system generates items to users, then the further recommendation model is trained with the data that users' feedback to the items. Such self-reinforcing pattern can cause data bias problems. There are several debiasing methods, inverse-propensity-scoring (IPS) is a practical one for industry product. Since it is relatively easy to reweight training samples, and ameliorate the distribution shift problem. However,because of deterministic policy problem and confoundings in real-world data, it is hard to predict propensity score accurately. Inspired by the sample reweight work for robust deep learning, we propose a novel influence function based method for recommendation modeling, and analyze how the influence function corrects the bias. In the experiments, our proposed method achieves better performance against the state-of-the-art approaches.

Click-through rate (CTR) prediction is the core problem of building advertising systems. Most existing state-of-the-art approaches model CTR prediction as binary classification problems, where displayed events with and without click feedbacks are respectively considered as positive and negative instances for training and offline validation. However, due to the selection mechanism applied in most advertising systems, a selection bias exists between distributions of displayed and non-displayed events. Conventional CTR models ignoring the bias may have inaccurate predictions and cause a loss of the revenue. To alleviate the bias, we need to conduct counterfactual learning by considering not only displayed events but also non-displayed events. In this paper, through a review of existing approaches of counterfactual learning, we point out some difficulties for applying these approaches for CTR prediction in a real-world advertising system. To overcome these difficulties, we propose a novel framework for counterfactual CTR prediction. In experiments, we compare our proposed framework against state-of-the-art conventional CTR models and existing counterfactual learning approaches. Experimental results show significant improvements.

Abstract Recently, cloud computing has drawn much attention from research and industrial communities. With adoption of cloud computing, data providers can reap huge economic benefits by outsourcing their data to cloud. Because of serious privacy concerns, sensitive data should be encrypted before being outsourced. In order to share the encrypted and outsourced data, data providers need to distribute different keys to legal users. Thus, in the scenario of multiple data providers, a legal user may receive lots of keys from different providers and suffer from heavy burden from key management. In this paper, we proposed an identity‐based key management scheme, which can let a legal user only maintain a single root key regardless of the number of data providers. Based on the root key, a legal user can derive different resource keys to decrypt ciphertexts from different data providers. With the help of outsourcing decryption technique, a legal user could reduce the decryption overhead by outsourcing some decryption work to cloud. We proofed the correctness and analyzed the security of the proposed scheme, further implemented the proposed scheme, and conducted extensive experiments. The results demonstrate the efficiency and practicality of our scheme. Copyright © 2016 John Wiley & Sons, Ltd.

As cloud computing becomes prevalent, many data owners outsource their large number of graph data into the cloud for cost savings. The cloud may not be fully trusted and usually likes to learn owners' data. To protect the security of graph data, sensitive data have to be stored in an encrypted form. Neighbor queries are the most essential operations, many other operations can be built based on the neighbor queries. Ranked neighbor queries can make the queries very flexible, a query user can get ranked search results. When the graph data are encrypted, the operations over graph data are very inconvenient. Ranked neighbor queries over encrypted graph data are difficult to directly achieve. In this paper, we propose a solution to perform privacy-preserving ranked neighbor query over encrypted graph data in the cloud, which is named as PRNQ. Our work utilizes ranking function and encryption mechanism to perform ranked neighbor query. The query users can obtain the most wanted search results by this solution. Our paper proposes a design scheme for the weighted directed graph and proves the security and correctness. Experimental results with a real graph data set demonstrate the validity and efficiency of our scheme. Copyright © 2016 John Wiley & Sons, Ltd.

Datasets containing individuals’ information (stored in back-end databases) are often published and shared in social networks. The disclosure of sensitive individuals’ information in social networks is potentially a serious problem. When an attacker studies a published table in a social network, the attacker could infer valuable information of individuals if the attacker learnt some sensitive information of other related individuals from other sources which are different from the published table. This type of attack is referred to as corruption attack. Existing privacy-preserving data publication (PPDP) approaches have been developed against corruption attacks, however, they could cause severe information loss, and reduce the usefulness of the published data. In addition, PPDP models based on ℓ-diversity and its variants may lead to individual sensitive information disclosure. Motivated by providing a solution to overcome these drawbacks, an independent ℓ-diversity principle is proposed in this study. Based on this principle a PPDP model is presented. The model could prevent attacks from attackers who have known data publishing algorithms and have the corruption abilities. A new data utility measurement global loss penalty is also proposed in this study. Related algorithms to our approach have been developed and implemented. Extensive experiments have been performed and comparisons with other related methods have been made. The results have shown the effectiveness of our approaches. It has been noted that when compared with ℓ-diversity model and its variant models, our model could resist corruption attacks more effectively; furthermore, when compared with other solutions against corruption attacks, our method would result in less information loss.

Abstract Private information retrieval (PIR) is a technique for clients to retrieve data from database while hiding the identity of data. Existing PIR schemes are impractical because of the expensive cost for complete privacy. In this paper, we propose a flexible, distributed scheme for PIR. First, a new protocol row‐based computational PIR (Robac‐PIR) is designed to make a trade‐off between privacy and efficiency. Second, we apply Robac‐PIR in a distributed system to further lower the computational cost. We implement the scheme, and the results show it has a good performance and can also provide the trade‐off. Copyright © 2014 John Wiley & Sons, Ltd.

The disclosure of individuals' sensitive values is a potentially serious problem in privacy preserving data publishing. The popular ℓ-diversity model is able to prevent victim' sensitive value from being revealed with a certainty higher than 1/ℓ. But, the anonymized tables with ℓ-diversity generated by the methods, such as generalization, anatomy and slicing, are all vulnerable to the attacker with some individuals' sensitive values. In this paper we propose a new approach to publish anonymized table with ℓ-diversity against the attacker having publishing algorithm and some individuals' sensitive values. Such an approach replaces the sensitive attribute (SA) value of each record with an values set consisted of the real SA value and several random noise values. The method can be applied to protect numerical and nominal SA, existing additive noise only is used to protect the numerical SA of anonymized table. Comparing with randomization, our method retains the frequency distribution of SA values of original data and retains more correlations (between SA and other attributes). Experimental evaluation shows that our method preserves more correlations than that of generalization, slicing and anatomy.

While outsourcing data to cloud, security and efficiency issues should be taken into account. However, it is very challenging to design a secure and efficient mechanism supporting authorization updates. In this paper, we aim to provide a mechanism supporting authorization updates which only incurs a lightweight cost of authorization updates and meanwhile supports a high level of security. This mechanism is consisted of two encryption schemes performed in different layers. The inner-layer encryption scheme is performed on the original plaintext and the generated cipher text is called inner-layer cipher text, while a part of the inner-layer cipher text is encrypted by the outer-layer encryption scheme to generate cipher text, called out-layer cipher text. These two encryption schemes are both performed by data owner. The inner-layer encryption realizes the initial authorization policy, while the outer-layer encryption reflects the updated authorization policy. We implement the proposed mechanism and conduct extensive experiments. The experimental results demonstrate that the proposed mechanism outperforms previous existing approaches, e.g. single-layer encryption and double-layer encryption.

ABSTRACT As cloud servers may not be trusted, sensitive data have to be transmitted and stored in an encrypted form. Major challenges for users are from the management (storage, update, protection, backup, and recoverability) of keys that can help users to decrypt authorized data available on the servers. In this paper, we propose a versatile approach for extremely lightweight key management, which is one of the most basic security tasks in cloud systems. In the multiple data owners scenario, each user only needs to manage a single key by our approach. With the help of the single key and a set of public information stored on the server, users can decrypt all authorized data from different data owners. Specifically, our paper proposes a novel access control model, proves the correctness and security, and analyzes the complexity of the model. Experimental results show that our approach significantly outperforms the single‐layer derivation encryption and double‐layer derivation encryption on the lightweight performance. Copyright © 2013 John Wiley & Sons, Ltd.

Fine-grained access control (FGAC) must be supported by relational databases to satisfy the requirements of privacy preserving and Internet-based applications. Though much work on FGAC models has been conducted, there are still a number of ongoing problems. We propose a new FGAC model which supports the specification of open access control policies as well as closed access control policies in relational databases. The negative authorization is supported, which allows the security administrator to specify what data should not be accessed by certain users. Moreover, multiple policies defined to regulate user access together are also supported. The definition and combination algorithm of multiple policies are thus provided. Finally, we implement the proposed FGAC model as a component of the database management system (DBMS) and evaluate its performance. The performance results show that the proposed model is feasible.

Data recovery techniques for malicious transactions are increasingly becoming an important issue since the security for DBMSs are mainly prevention based, and they cannot defend systems from unknown attacks. Survivability and availability are essential for modern DBMSs, which require the database provide continuous services in the period of recovery, namely dynamic recovery. In this paper, we presented a data recovery model and introduce extended read-write dependency and phantoms dependency to the model. A fine grained transaction log is proposed for data recovery. The log records all the data items of the read and update-involved operations for the committed transactions, and even extracts data items read by the subqueries in the SQL statements. Based on the log, we develop a dynamic recovery system to implement the data recovery model. The system could provide continuous services while the recovery is processing. Experiments based on TPC-W benchmark show that the dynamic recovery system is high-efficient and reliable.

The fine-grained access control (FGAC) is important to Web applications. However, it can be circumvented by indirect access such as inferring sensitive data from insensitive data. In the previous FGAC studies, the inference information leakage is not well controlled. In addition, the soundness property, which requires that the results of a query under the FGAC should be the subset of the results of the query without the FGAC over the same database states, can not be held for all SQL statements. Moreover, without soundness information leakage would occur in databases in some situations. In this paper, we divide the approaches which implement FGAC with query modification into two types: the first, the FGAC policy is executed over the results obtained from the query issued by a user; the second, the query issued by a user is executed over the results obtained from the FGAC policy. Then we introduce three types of information leakages of FGAC, which is implemented by using the first approach, and analyze why the existing approaches can not hold soundness for all SQL statements. A novel approach is proposed to control the information leakages and satisfy the soundness property for all SQL statements. Then we implement our approach in DM DBMS with query modification and analyze the test results.