The integration of large language models (LLMs) with external content has enabled applications such as Microsoft Copilot but also introduced vulnerabilities to indirect prompt injection attacks. In these attacks, malicious instructions embedded within external content can manipulate LLM outputs, causing deviations from user expectations. To address this critical yet under-explored issue, we introduce the first benchmark for bindirect prompt injection attacks, named BIPIA, to assess the risk of such vulnerabilities. Using BIPIA, we evaluate existing LLMs and find them universally vulnerable. Our analysis identifies two key factors contributing to their success: LLMs' inability to distinguish between informational context and actionable instructions, and their lack of awareness in avoiding the execution of instructions within external content. Based on these findings, we propose two novel defense mechanisms -- boundary awareness and explicit reminder -- to address these vulnerabilities in both black-box and white-box settings. Extensive experiments demonstrate that our black-box defense provides substantial mitigation, while our white-box defense reduces the attack success rate to near-zero levels, all while preserving the output quality of LLMs. We hope this work inspires further research into securing LLM applications and fostering their safe and reliable use. Our code is available at https://github.com/microsoft/BIPIA.

In educational data mining, concept prerequisite relations extraction determines which concepts need to be learned before learning another concept. It plays a crucial role in pedagogical practices, such as learning path planning and curriculum design. Deep neural networks, especially graph neural networks, have recently made significant strides in concept prerequisite relations extraction. However, existing methods face two primary limitations. (1) Methods with better performance construct heterogeneous complete graphs, leading to higher model complexity and training cost. Meanwhile, the performance of low-complexity methods is inferior to the former. (2) A disregard for temporal context, essential for learning, limits both the performance and the application of these methods. To address these issues, we propose a novel graph-based approach, called Learning-path based Concept Prerequisite Relations Extraction (LCPRE). LCPRE constructs a lightweight sparse graph in a simple manner, which reduces complexity from quadratic to linear and captures the temporal feature through learning-path, a comprehensible learning approach from one concept to another. Experimental results on three benchmark datasets demonstrate that LCPRE outperforms existing methods, establishing a new state-of-the-art in concept prerequisite relations extraction.

News recommendation is essential for personalized news distribution. Federated news recommendation, which enables collaborative model learning from multiple clients without sharing their raw data, is a promising approach for preserving users' privacy. However, the security of federated news recommendation is still unclear. In this paper, we study this problem by proposing an untargeted attack on federated news recommendation called UA-FedRec. By exploiting the prior knowledge of news recommendation and federated learning, UA-FedRec can effectively degrade the model performance with a small percentage of malicious clients. First, the effectiveness of news recommendation highly depends on user modeling and news modeling. We design a news similarity perturbation method to make representations of similar news farther and those of dissimilar news closer to interrupt news modeling, and propose a user model perturbation method to make malicious user updates in opposite directions of benign updates to interrupt user modeling. Second, updates from different clients are typically aggregated with a weighted average based on their sample sizes. We propose a quantity perturbation method to enlarge sample sizes of malicious clients in a reasonable range to amplify the impact of malicious updates. Extensive experiments on two real-world datasets show that UA-FedRec can effectively degrade the accuracy of existing federated news recommendation methods, even when defense is applied. Our study reveals a critical security issue in existing federated news recommendation systems and calls for research efforts to address the issue. Our code is available at https://github.com/yjw1029/UA-FedRec.

Candidate generation task requires that candidates related to user interests need to be extracted in realtime. Previous works usually transform a user's behavior sequence to a unified embedding, which can not reflect the user's multiple interests. Some recent works like Comirec and Octopus use multi-channel structures to capture users' diverse interests. They cluster users' historical behaviors into several groups, claiming that one group represents one interest. However, these methods have some limitations. First, an item may correspond to multiple interests of users, thereby simply allocating it to just one interest group will make the modeling of users' interests coarse-grained and inaccurate. Second, explaining user interests at the level of items is rather vague and not convincing. In this paper, we propose a Knowledge Enhanced Multi-Interest Network: KEMI, which exploits knowledge graphs to help learn users' diverse interest representations via heterogeneous graph neural networks (HGNNs) and a novel dual memory network. Specifically, we use HGNNs to capture the semantic representation of knowledge entities and a novel dual memory network to learn a user's diverse interests from his behavior sequence. Through memory slots of the user memory network and the item memory network, we can learn multiple interests for each user and each item. Meanwhile, by binding the entities to the channels of memory networks, we enable it to be explained from the perspective of the knowledge graph, which enhances the interpretability and understanding of user interests. We conduct extensive experiments on two industrial and publicly available datasets. Experimental results demonstrate that our model achieves significant improvements over state-of-the-art baseline models.

News recommendation systems play a key role in online news reading service. Knowledge graphs (KG), which contain comprehensive structural knowledge, are well known for their potential to enhance both accuracy and explainability. While existing works intensively study using KG to improve news recommendation accuracy, using KG for news recommendation reasoning has not been fully explored. A few works such as KPRN [18], [22] and ADAC [25] have discussed knowledge reasoning in some other recommendation domains such as music or movie, but their methods are not practical for the news. How to make reasoning scalable to generic KGs, easy to deploy for real-time serving and meanwhile elastic for both recall and ranking stages remains an open question.

News articles usually contain knowledge entities such as celebrities or organizations. Important entities in articles carry key messages and help to understand the content in a more direct way. An industrial news recommender system contains various key applications, such as personalized recommendation, item-to-item recommendation, news category classification, news popularity prediction and local news detection. We find that incorporating knowledge entities for better document understanding benefits these applications consistently. However, existing document understanding models either represent news articles without considering knowledge entities (e.g., BERT) or rely on a specific type of text encoding model (e.g., DKN) so that the generalization ability and efficiency is compromised. In this paper, we propose KRED, which is a fast and effective model to enhance arbitrary document representation with a knowledge graph. KRED first enriches entities’ embeddings by attentively aggregating information from their neighborhood in the knowledge graph. Then a context embedding layer is applied to annotate the dynamic context of different entities such as frequency, category and position. Finally, an information distillation layer aggregates the entity embeddings under the guidance of the original document representation and transforms the document vector into a new one. We advocate to optimize the model with a multi-task framework, so that different news recommendation applications can be united and useful information can be shared across different tasks. Experiments on a real-world Microsoft News dataset demonstrate that KRED greatly benefits a variety of news recommendation applications.

Combinatorial features are essential for the success of many commercial models. Manually crafting these features usually comes with high cost due to the variety, volume and velocity of raw data in web-scale systems. Factorization based models, which measure interactions in terms of vector product, can learn patterns of combinatorial features automatically and generalize to unseen features as well. With the great success of deep neural networks (DNNs) in various fields, recently researchers have proposed several DNN-based factorization model to learn both low- and high-order feature interactions. Despite the powerful ability of learning an arbitrary function from data, plain DNNs generate feature interactions implicitly and at the bit-wise level. In this paper, we propose a novel Compressed Interaction Network (CIN), which aims to generate feature interactions in an explicit fashion and at the vector-wise level. We show that the CIN share some functionalities with convolutional neural networks (CNNs) and recurrent neural networks (RNNs). We further combine a CIN and a classical DNN into one unified model, and named this new model eXtreme Deep Factorization Machine (xDeepFM). On one hand, the xDeepFM is able to learn certain bounded-degree feature interactions explicitly; on the other hand, it can learn arbitrary low- and high-order feature interactions implicitly. We conduct comprehensive experiments on three real-world datasets. Our results demonstrate that xDeepFM outperforms state-of-the-art models. We have released the source code of xDeepFM at https://github.com/Leavingseason/xDeepFM.

Microblogging Web sites, such as Twitter and Sina Weibo, have become popular platforms for socializing and sharing information in recent years. Spammers have also discovered this new opportunity to unfairly overpower normal users with unsolicited content, namely social spams. Although it is intuitive for everyone to follow legitimate users, recent studies show that both legitimate users and spammers follow spammers for different reasons. Evidence of users seeking spammers on purpose is also observed. We regard this behavior as useful information for spammer detection. In this article, we approach the problem of spammer detection by leveraging the “carefulness” of users, which indicates how careful a user is when she is about to follow a potential spammer. We propose a framework to measure the carefulness and develop a supervised learning algorithm to estimate it based on known spammers and legitimate users. We illustrate how the robustness of the detection algorithms can be improved with aid of the proposed measure. Evaluation on two real datasets from Sina Weibo and Twitter with millions of users are performed, as well as an online test on Sina Weibo. The results show that our approach indeed captures the carefulness, and it is effective for detecting spammers. In addition, we find that our measure is also beneficial for other applications, such as link prediction.

Point-of-Interest (POI) recommendation has become an important means to help people discover attractive locations. However, extreme sparsity of user-POI matrices creates a severe challenge. To cope with this challenge, viewing mobility records on location-based social networks (LBSNs) as implicit feedback for POI recommendation, we first propose to exploit weighted matrix factorization for this task since it usually serves collaborative filtering with implicit feedback better. Besides, researchers have recently discovered a spatial clustering phenomenon in human mobility behavior on the LBSNs, i.e., individual visiting locations tend to cluster together, and also demonstrated its effectiveness in POI recommendation, thus we incorporate it into the factorization model. Particularly, we augment users' and POIs' latent factors in the factorization model with activity area vectors of users and influence area vectors of POIs, respectively. Based on such an augmented model, we not only capture the spatial clustering phenomenon in terms of two-dimensional kernel density estimation, but we also explain why the introduction of such a phenomenon into matrix factorization helps to deal with the challenge from matrix sparsity. We then evaluate the proposed algorithm on a large-scale LBSN dataset. The results indicate that weighted matrix factorization is superior to other forms of factorization models and that incorporating the spatial clustering phenomenon into matrix factorization improves recommendation performance.

Smart card transactions capture rich information of human mobility and urban dynamics, therefore are of particular interest to urban planners and location-based service providers. However, since most transaction systems are only designated for billing purpose, typically, fine-grained location information, such as the exact boarding and alighting stops of a bus trip, is only partially or not available at all, which blocks deep exploitation of this rich and valuable data at individual level. This paper presents a "space alignment" framework to reconstruct individual mobility history from a large-scale smart card transaction dataset pertaining to a metropolitan city. Specifically, we show that by delicately aligning the monetary space and geospatial space with the temporal space, we are able to extrapolate a series of critical domain specific constraints. Later, these constraints are naturally incorporated into a semi-supervised conditional random field to infer the exact boarding and alighting stops of all transit routes with a surprisingly high accuracy, e.g., given only 10% trips with known alighting/boarding stops, we successfully inferred more than 78% alighting and boarding stops from all unlabeled trips. In addition, we demonstrated that the smart card data enriched by the proposed approach dramatically improved the performance of a conventional method for identifying users' home and work places (with 88% improvement on home detection and 35% improvement on work place detection). The proposed method offers the possibility to mine individual mobility from common public transit transactions, and showcases how uncertain data can be leveraged with domain knowledge and constraints, to support cross-application data mining tasks.

The problem of computing minimum distance of moving objects in large-scale road networks has caught many researchers` attention in recent years. One of solutions is to reduce the complexity by selecting representative node for each node in graph, in which the mininum distance of two nodes can be obtained by computing the distance between their representative nodes. The distance between any two representative nodes is pre-computed. In this paper we propose an effective algorithm to select and allocate a fixed number of representative nodes in road networks, then analyse the relationship between the size of representative node set and the approximation distance error of node pairs. We evaluate our approach on a real data set. The evaluation result shows the size of representative nodes set affects the approximation distance error of node pairs, and the appropriate choice for size of representative nodes set could improve the performance of precomputation and make the result more accurate and reliable.

This paper presents a smart driving direction system leveraging the intelligence of experienced drivers. In this system, GPS-equipped taxis are employed as mobile sensors probing the traffic rhythm of a city and taxi drivers' intelligence in choosing driving directions in the physical world. We propose a time-dependent landmark graph to model the dynamic traffic pattern as well as the intelligence of experienced drivers so as to provide a user with the practically fastest route to a given destination at a given departure time. Then, a Variance-Entropy-Based Clustering approach is devised to estimate the distribution of travel time between two landmarks in different time slots. Based on this graph, we design a two-stage routing algorithm to compute the practically fastest and customized route for end users. We build our system based on a real-world trajectory data set generated by over 33,000 taxis in a period of three months, and evaluate the system by conducting both synthetic experiments and in-the-field evaluations. As a result, 60-70 percent of the routes suggested by our method are faster than the competing methods, and 20 percent of the routes share the same results. On average, 50 percent of our routes are at least 20 percent faster than the competing approaches.

This paper presents a Cloud-based system computing customized and practically fast driving routes for an end user using (historical and real-time) traffic conditions and driver behavior. In this system, GPS-equipped taxicabs are employed as mobile sensors constantly probing the traffic rhythm of a city and taxi drivers' intelligence in choosing driving directions in the physical world. Meanwhile, a Cloud aggregates and mines the information from these taxis and other sources from the Internet, like Web maps and weather forecast. The Cloud builds a model incorporating day of the week, time of day, weather conditions, and individual driving strategies (both of the taxi drivers and of the end user for whom the route is being computed). Using this model, our system predicts the traffic conditions of a future time (when the computed route is actually driven) and performs a self-adaptive driving direction service for a particular user. This service gradually learns a user's driving behavior from the user's GPS logs and customizes the fastest route for the user with the help of the Cloud. We evaluate our service using a real-world dataset generated by over 33,000 taxis over a period of 3 months in Beijing. As a result, our service accurately estimates the travel time of a route for a user; hence finding the fastest route customized for the user.

GPS-equipped taxis can be regarded as mobile sensors probing traffic flows on road surfaces, and taxi drivers are usually experienced in finding the fastest (quickest) route to a destination based on their knowledge. In this paper, we mine smart driving directions from the historical GPS trajectories of a large number of taxis, and provide a user with the practically fastest route to a given destination at a given departure time. In our approach, we propose a time-dependent landmark graph, where a node (landmark) is a road segment frequently traversed by taxis, to model the intelligence of taxi drivers and the properties of dynamic road networks. Then, a Variance-Entropy-Based Clustering approach is devised to estimate the distribution of travel time between two landmarks in different time slots. Based on this graph, we design a two-stage routing algorithm to compute the practically fastest route. We build our system based on a real-world trajectory dataset generated by over 33,000 taxis in a period of 3 months, and evaluate the system by conducting both synthetic experiments and in-the-field evaluations. As a result, 60-70% of the routes suggested by our method are faster than the competing methods, and 20% of the routes share the same results. On average, 50% of our routes are at least 20% faster than the competing approaches.

The emerging location-detection devices together with ubiquitous connectivity have enabled a large variety of location-based services (LBS). Unfortunately, LBS may threaten the users' privacy. K-anonymity cloaking the user location to K-anonymizing spatial region (K-ASR) has been extensively studied to protect privacy in LBS. Traditional K-anonymity method needs complex query processing algorithms at the server side. SpaceTwist rectifies the above shortcoming of traditional K-anonymity since it only requires incremental nearest neighbor (INN) queries processing techniques at the server side. However, Space Twist may fail since it cannot guarantee K-anonymity. In this paper, our proposed framework, called KAWCR (K-anonymity Without Cloaked Region), rectifies the shortcomings and retains the advantages of the above two techniques. KAWCR only needs the server to process INN queries and can guarantee that the users issuing the query is indistinguishable from at least K-1 other users. The extensive experimental results show that the communication cost of KAWCR for kNN queries is lower than that of both traditional K-anonymity and SpaceTwist.

Matching a raw GPS trajectory to roads on a digital map is often referred to as the Map Matching problem. However, the occurrence of the low-sampling-rate trajectories (e.g. one point per 2 minutes) has brought lots of challenges to existing map matching algorithms. To address this problem, we propose an Interactive Voting-based Map Matching (IVMM) algorithm based on the following three insights: 1) The position context of a GPS point as well as the topological information of road networks, 2) the mutual influence between GPS points (i.e., the matching result of a point references the positions of its neighbors; in turn, when matching its neighbors, the position of this point will also be referenced), and 3) the strength of the mutual influence weighted by the distance between GPS points (i.e., the farther distance is the weaker influence exists). In this approach, we do not only consider the spatial and temporal information of a GPS trajectory but also devise a voting-based strategy to model the weighted mutual influences between GPS points. We evaluate our IVMM algorithm based on a user labeled real trajectory dataset. As a result, the IVMM algorithm outperforms the related method (ST-Matching algorithm).