Given a social network G and a constant $k$, the influence maximization problem asks for k nodes in G that (directly and indirectly) influence the largest number of nodes under a pre-defined diffusion model. This problem finds important applications in viral marketing, and has been extensively studied in the literature. Existing algorithms for influence maximization, however, either trade approximation guarantees for practical efficiency, or vice versa. In particular, among the algorithms that achieve constant factor approximations under the prominent independent cascade (IC) model or linear threshold (LT) model, none can handle a million-node graph without incurring prohibitive overheads. This paper presents TIM, an algorithm that aims to bridge the theory and practice in influence maximization. On the theory side, we show that TIM runs in O((k+ l) (n+m) log n/ε2) expected time and returns a (1-1/e-ε)-approximate solution with at least 1 - n-l probability. The time complexity of TIM is near-optimal under the IC model, as it is only a log n factor larger than the Ω(m + n) lower-bound established in previous work (for fixed k, l, and ε). Moreover, TIM supports the triggering model, which is a general diffusion model that includes both IC and LT as special cases. On the practice side, TIM incorporates novel heuristics that significantly improve its empirical efficiency without compromising its asymptotic performance. We experimentally evaluate TIM with the largest datasets ever tested in the literature, and show that it outperforms the state-of-the-art solutions (with approximation guarantees) by up to four orders of magnitude in terms of running time. In particular, when k = 50, ε = 0.2, and l = 1, TIM requires less than one hour on a commodity machine to process a network with 41.6 million nodes and 1.4 billion edges. This demonstrates that influence maximization algorithms can be made practical while still offering strong theoretical guarantees.

We study generalization for preserving privacy in publication of sensitive data. The existing methods focus on a universal approach that exerts the same amount of preservation for all persons, with- out catering for their concrete needs. The consequence is that we may be offering insufficient protection to a subset of people, while applying excessive privacy control to another subset. Motivated by this, we present a new generalization framework based on the concept of personalized anonymity. Our technique performs the minimum generalization for satisfying everybody's requirements, and thus, retains the largest amount of information from the microdata. We carry out a careful theoretical study that leads to valuable insight into the behavior of alternative solutions. In particular, our analysis mathematically reveals the circumstances where the previous work fails to protect privacy, and establishes the superiority of the proposed solutions. The theoretical findings are verified with extensive experiments.

Privacy-preserving data publishing has attracted considerable research interest in recent years. Among the existing solutions, ∈-differential privacy provides the strongest privacy guarantee. Existing data publishing methods that achieve ∈-differential privacy, however, offer little data utility. In particular, if the output data set is used to answer count queries, the noise in the query answers can be proportional to the number of tuples in the data, which renders the results useless. In this paper, we develop a data publishing technique that ensures ∈-differential privacy while providing accurate answers for range-count queries, i.e., count queries where the predicate on each attribute is a range. The core of our solution is a framework that applies wavelet transforms on the data before adding noise to it. We present instantiations of the proposed framework for both ordinal and nominal data, and we provide a theoretical analysis on their privacy and utility guarantees. In an extensive experimental study on both real and synthetic data, we show the effectiveness and efficiency of our solution.

Recommender systems are collecting and analyzing user data to provide better user experience. However, several privacy concerns have been raised when a recommender knows user's set of items or their ratings. A number of solutions have been suggested to improve privacy of legacy recommender systems, but the existing solutions in the literature can protect either items or ratings only. In this paper, we propose a recommender system that protects both user's items and ratings. We develop novel matrix factorization algorithms under local differential privacy (LDP). In a recommender system with LDP, individual users randomize their data themselves to satisfy differential privacy and send the perturbed data to the recommender. Then, the recommender computes aggregates of the perturbed data. This framework ensures that both user's items and ratings remain private from the recommender. However, applying LDP to matrix factorization typically raises utility issues with high dimensionality and iterative algorithms. To tackle these technical challenges, we adopt dimensionality reduction technique and a sampling-based binary mechanism. We introduce a factor that stabilizes the perturbed gradients. With MovieLens and LibimSeTi datasets, we evaluate accuracy of our recommender system and demonstrate that our algorithm performs better than the existing differentially private gradient descent algorithm for matrix factorization under stronger privacy requirements.

Numerous applications require continuous publication of statistics or monitoring purposes, such as real-time traffic analysis, timely disease outbreak discovery, and social trends observation. These statistics may be derived from sensitive user data and, hence, necessitate privacy preservation. A notable paradigm for offering strong privacy guarantees in statistics publishing is ε-differential privacy. However, there is limited literature that adapts this concept to settings where the statistics are computed over an infinite stream of "events" (i.e., data items generated by the users), and published periodically. These works aim at hiding a single event over the entire stream. We argue that, in most practical scenarios, sensitive information is revealed from multiple events occurring at contiguous time instances. Towards this end, we put forth the novel notion of w - event privacy over infinite streams, which protects any event sequence occurring in w successive time instants. We first formulate our privacy concept, motivate its importance, and introduce a methodology for achieving it. We next design two instantiations, whose utility is independent of the stream length. Finally, we confirm the practicality of our solutions experimenting with real data.