The k-center clustering problem is of fundamental importance for a broad range of machine learning and data science applications. In this paper, we study the deletion-robust version of the problem. Specifically, we aim to extract a small subset of a given data set, referred to as a coreset, that contains a provably good set of k centers even after an adversary deletes up to z arbitrarily chosen points from the data set. We propose a 4-approximation algorithm that provides a coreset of size O(kz). To our knowledge, this is the first algorithm for deletion-robust k-center clustering with a theoretical guarantee. Moreover, we accompany our theoretical results with extensive experiments, demonstrating that our algorithm achieves significantly better robustness than non-trivial baselines against three heuristic gray-box and white-box adversarial deletion attacks.

Retraining a machine learning (ML) model is essential for maintaining its performance as the data change over time. However, retraining is also costly, as it typically requires re-processing the entire dataset. As a result, a trade-off arises: on the one hand, retraining an ML model too frequently incurs unnecessary computing costs; on the other hand, not retraining frequently enough leads to stale ML models and incurs a cost in loss of accuracy. To resolve this trade-off, we envision ML systems that make automated and cost-optimal decisions about when to retrain an ML model. In this work, we study the decision problem of whether to retrain or keep an existing ML model based on the data, the model, and the predictive queries answered by the model. Crucially, we consider the costs associated with each decision and aim to optimize the trade-off. Our main contribution is a Cost-Aware Retraining Algorithm, CARA, which optimizes the trade-off over streams of data and queries. To explore the performance of CARA, we first analyze synthetic datasets and demonstrate that CARA can adapt to different data drifts and retraining costs while performing similarly to an optimal retrospective algorithm. Subsequently, we experiment with real-world datasets and demonstrate that CARA has better accuracy than drift detection baselines while making fewer retraining decisions, thus incurring lower total costs.

Recommender systems typically suggest to users content similar to what they consumed in the past. A user, if happening to be exposed to strongly polarized content, might be steered towards more and more radicalized content by subsequent recommendations, eventually being trapped in what we call a "radicalization pathway". In this paper, we investigate how to mitigate radicalization pathways using a graph-based approach. We model the set of recommendations in a what-to-watch-next (W2W) recommender as a directed graph, where nodes correspond to content items, links to recommendations, and paths to possible user sessions. We measure the segregation score of a node representing radicalized content as the expected length of a random walk from that node to any node representing non-radicalized content. A high segregation score thus implies a larger chance of getting users trapped in radicalization pathways. We aim to reduce the prevalence of radicalization pathways by selecting a small number of edges to rewire, so as to minimize the maximum of segregation scores among all radicalized nodes while maintaining the relevance of recommendations. We propose an efficient yet effective greedy heuristic based on the absorbing random walk theory for the rewiring problem. Our experiments on real-world datasets confirm the effectiveness of our proposal.

Diversity maximization aims to select a diverse and representative subset of items from a large dataset. It is a fundamental optimization task that finds applications in data summarization, feature selection, web search, recommender systems, and elsewhere. However, in a setting where data items are associated with different groups according to sensitive attributes like sex or race, it is possible that algorithmic solutions for this task, if left unchecked, will under- or over- represent some of the groups. Therefore, we are motivated to address the problem of max-min diversification with fairness constraints, aiming to select k items to maximize the minimum distance between any pair of selected items while ensuring that the number of items selected from each group falls within predefined lower and upper bounds. In this work, we propose an exact algorithm based on integer linear programming that is suitable for small datasets as well as a -approximation algorithm for any parameter ɛ ∊ (0,1) that scales to large datasets. Extensive experiments on real-world datasets demonstrate the superior performance of our proposed algorithms over existing ones.

Graph summarization via node grouping is a popular method to build concise graph representations by grouping nodes from the original graph into supernodes and encoding edges into superedges such that the loss of adjacency information is minimized. Such summaries have immense applications in large-scale graph analytics due to their small size and high query processing efficiency. In this paper, we reformulate the loss minimization problem for summarization into an equivalent integer maximization problem. By initially allowing relaxed (fractional) solutions for integer maximization, we analytically expose the underlying connections to the spectral properties of the adjacency matrix. Consequently, we design an algorithm called SpecSumm that consists of two phases. In the first phase, motivated by spectral graph theory, we apply k-means clustering on the k largest (in magnitude) eigenvectors of the adjacency matrix to assign nodes to supernodes. In the second phase, we propose a greedy heuristic that updates the initial assignment to further improve summary quality. Finally, via extensive experiments on 11 datasets, we show that SpecSumm efficiently produces high-quality summaries compared to state-of-the-art summarization algorithms and scales to graphs with millions of nodes.

Linear classifiers are well-known to be vulnerable to adversarial attacks: they may predict incorrect labels for input data that are adversarially modified with small perturbations. However, this phenomenon has not been properly understood in the context of sketch-based linear classifiers, typically used in memory-constrained paradigms, which rely on random projections of the features for model compression. In this paper, we propose novel Fast-Gradient-Sign Method (FGSM) attacks for sketched classifiers in full, partial, and black-box information settings with regards to their internal parameters. We perform extensive experiments on the MNIST dataset to characterize their robustness as a function of perturbation budget. Our results suggest that, in the full-information setting, these classifiers are less accurate on unaltered input than their uncompressed counterparts but just as susceptible to adversarial attacks. But in more realistic partial and black-box information settings, sketching improves robustness while having lower memory footprint.

Bayesian networks are popular probabilistic models that capture the conditional dependencies among a set of variables. Inference in Bayesian networks is a fundamental task for answering probabilistic queries over a subset of variables in the data. However, exact inference in Bayesian networks is NP-hard, which has prompted the development of many practical inference methods. In this paper, we focus on improving the performance of the junction-tree algorithm, a well-known method for exact inference in Bayesian networks. In particular, we seek to leverage information in the workload of probabilistic queries to obtain an optimal workload-aware materialization of junction trees, with the aim to accelerate the processing of inference queries. We devise an optimal pseudo-polynomial algorithm to tackle this problem and discuss approximation schemes. Compared to state-of-the-art approaches for efficient processing of inference queries via junction trees, our methods are the first to exploit the information provided in query workloads. Our experimentation on several real-world Bayesian networks confirms the effectiveness of our techniques in speeding-up query processing.

Diversity maximization is a fundamental problem with wide applications in data summarization, web search, and recommender systems. Given a set$X$of$n$elements, it asks to select a subset$S$of$k\ll n$elements with maximum diversity, as quantified by the dissimilarities among the elements in S. In this paper, we focus on the diversity maximization problem with fairness constraints in the streaming setting. Specifically, we consider the max-min diversity objective, which selects a subset$S$that maximizes the minimum distance (dissimilarity) between any pair of distinct elements within it. Assuming that the set$X$is partitioned into$m$disjoint groups by some sensitive attribute, e.g., sex or race, ensuring fairness requires that the selected subset$S$contains kielements from each group i є [1, m]. A streaming algorithm should process$X$sequentially in one pass and return a subset with maximum diversity while guaranteeing the fairness constraint. Although diversity maximization has been extensively studied, the only known algorithms that can work with the max-min diversity objective and fairness constraints are very inefficient for data streams. Since diversity maximization is NP-hard in general, we propose two approximation algorithms for fair diversity maximization in data streams, the first of which is$\frac{1-\varepsilon}{4}$-approximate and specific for m = 2, where є E (0,1), and the second of which achieves a$\frac{1-\varepsilon}{3m+2}$-approximation for an arbitrary$m$. Experimental results on real-world and synthetic datasets show that both algorithms provide solutions of comparable quality to the state-of-the-art algorithms while running several orders of magnitude faster in the streaming setting.

Recommender systems typically suggest to users content similar to what they consumed in the past. If a user happens to be exposed to strongly polarized content, she might subsequently receive recommendations which may steer her towards more and more radicalized content, eventually being trapped in what we call a “radicalization pathway”. In this paper, we study the problem of mitigating radicalization pathways using a graph-based approach. Specifically, we model the set of recommendations of a “what-to-watch-next” recommender as a d-regular directed graph where nodes correspond to content items, links to recommendations, and paths to possible user sessions.

Distance oracles answer shortest-path queries between any pair of nodes in a graph. They are often built using succinct graph representations such as spanners, sketches, and compressors to minimize oracle size and query answering latency. Node embeddings, in particular, offer graph representations that place adjacent nodes nearby each other in a low-rank space. However, their use in the design of distance oracles has not been sufficiently studied. In this paper, we empirically compare exact distance oracles constructed based on a variety of node embeddings and other succinct representations. We evaluate twelve such oracles along three measures of efficiency: construction time, memory requirements, and query-processing time over fourteen real datasets and four synthetic graphs. We show that distances between embedding vectors are excellent estimators of graph distances when graphs are well-structured, but less so for more unstructured graphs. Overall, our findings suggest that exact oracles based on embeddings can be constructed faster than multi-dimensional scaling (MDS) but slower than compressed adjacency indexes, require less memory than landmark oracles but more than sparsifiers or indexes, can answer queries faster than indexes but slower than MDS, and are exact more often with a smaller additive error than spanners (that have multiplicative error) while not being lossless like adjacency lists. Finally, while the exactness of such oracles is infeasible to maintain for huge graphs even under large amounts of resources, we empirically demonstrate that approximate oracles based on GOSH embeddings can efficiently scale to graphs of 100M+ nodes with only small additive errors in distance estimations.

Bayesian networks are general, well-studied probabilistic models that capture dependencies among a set of variables. Variable Elimination is a fundamental algorithm for probabilistic inference over Bayesian networks. In this paper, we propose a novel materialization method, which can lead to significant efficiency gains when processing inference queries using the Variable Elimination algorithm. In particular, we address the problem of choosing a set of intermediate results to precompute and materialize, so as to maximize the expected efficiency gain over a given query workload. For the problem we consider, we provide an optimal polynomial-time algorithm and discuss alternative methods. We validate our technique using real-world Bayesian networks. Our experimental results confirm that a modest amount of materialization can lead to significant improvements in the running time of queries, with an average gain of 70%, and reaching up to a gain of 99%, for a uniform workload of queries. Moreover, in comparison with existing junction tree methods that also rely on materialization, our approach achieves competitive efficiency during inference using significantly lighter materialization.

Coresets are succinct summaries of large datasets such that, for a given problem, the solution obtained from a coreset is provably competitive with the solution obtained from the full dataset. As such, coreset-based data summarization techniques have been successfully applied to various problems, e.g., geometric optimization, clustering, and approximate query processing, for scaling them up to massive data. In this paper, we study coresets for the maxima representation of multidimensional data: Given a set P of points in $ \mathbbR ^d $, where d is a small constant, and an error parameter $ \varepsilon \in (0,1) $, a subset $ Q \subseteq P $ is an $ \varepsilon $-coreset for the maxima representation of P iff the maximum of Q is an $ \varepsilon $-approximation of the maximum of P for any vector $ u \in \mathbbR ^d $, where the maximum is taken over the inner products between the set of points (P or Q) and u. We define a novel minimum $\varepsilon$-coreset problem that asks for an $\varepsilon$-coreset of the smallest size for the maxima representation of a point set. For the two-dimensional case, we develop an optimal polynomial-time algorithm for the minimum $ \varepsilon $-coreset problem by transforming it into the shortest-cycle problem in a directed graph. Then, we prove that this problem is NP-hard in three or higher dimensions and present polynomial-time approximation algorithms in an arbitrary fixed dimension. Finally, we provide extensive experimental results on both real and synthetic datasets to demonstrate the superior performance of our proposed algorithms.

We study the problem of extracting a small subset of representative items from a large data stream. In many data mining and machine learning applications such as social network analysis and recommender systems, this problem can be formulated as maximizing a monotone submodular function subject to a cardinality constraint k. In this work, we consider the setting where data items in the stream belong to one of several disjoint groups and investigate the optimization problem with an additional fairness constraint that limits selection to a given number of items from each group. We then propose efficient algorithms for the fairness-aware variant of the streaming submodular maximization problem. In particular, we first give a -approximation algorithm that requires passes over the stream for any constant ε > 0. Moreover, we give a single-pass streaming algorithm that has the same approximation ratio of when unlimited buffer sizes and post-processing time are permitted, and discuss how to adapt it to more practical settings where the buffer sizes are bounded. Finally, we demonstrate the efficiency and effectiveness of our proposed algorithms on two real-world applications, namely maximum coverage on large graphs and personalized recommendation.

Controversial issues often split the population into groups with opposing views. When such issues emerge on social media, we often observe the creation of "echo chambers," i.e., situations where like-minded people reinforce each other’s opinion, but do not get exposed to the views of the opposing side. In this paper we study algorithmic techniques for bridging these chambers, and thus reduce controversy. Specifically, we represent discussions as graphs, and cast our objective as an edge-recommendation problem. The goal of the recommendation is to reduce the controversy score of the graph, measured by a recently-developed metric based on random walks. At the same time, we take into account the acceptance probability of the recommended edges, which represent the probability that the recommended edges materialize in the graph.

In networking applications, one often wishes to obtain estimates about the number of objects at different parts of the network (e.g., the number of cars at an intersection of a road network or the number of packets expected to reach a node in a computer network) by monitoring the traffic in a small number of network nodes or edges. We formalize this task by defining the Markov Chain Monitoring problem. Given an initial distribution of items over the nodes of a Markov chain, we wish to estimate the distribution of items at subsequent times. We do this by asking a limited number of queries that retrieve, for example, how many items transitioned to a specific node or over a specific edge at a particular time. We consider different types of queries, each defining a different variant of the Markov Chain Monitoring. For each variant, we design efficient algorithms for choosing the queries that make our estimates as accurate as possible. In our experiments with synthetic and real datasets, we demonstrate the efficiency and the efficacy of our algorithms in a variety of settings.

Echo chambers, i.e., situations where one is exposed only to opinions that agree with their own, are an increasing concern for the political discourse in many democratic countries. This paper studies the phenomenon of political echo chambers on social media. We identify the two components in the phenomenon: the opinion that is shared, and the »chamber» (i.e., the social network) that allows the opinion to »echo» (i.e., be re-shared in the network) -- and examine closely at how these two components interact. We define a production and consumption measure for social-media users, which captures the political leaning of the content shared and received by them. By comparing the two, we find that Twitter users are, to a large degree, exposed to political opinions that agree with their own. We also find that users who try to bridge the echo chambers, by sharing content with diverse leaning, have to pay a »price of bipartisanship» in terms of their network centrality and content appreciation. In addition, we study the role of »gatekeepers,» users who consume content with diverse leaning but produce partisan content (with a single-sided leaning), in the formation of echo chambers. Finally, we apply these findings to the task of predicting partisans and gatekeepers from social and content features. While partisan users turn out relatively easy to identify, gatekeepers prove to be more challenging.

Digital location traces can help build insights about how citizens experience their cities, but also offer personalized products and experiences to them. Even as data abound, though, building an accurate picture about citizen whereabouts is not always straightforward, due to noisy or incomplete data.

Society is often polarized by controversial issues that split the population into groups with opposing views. When such issues emerge on social media, we often observe the creation of `echo chambers', i.e., situations where like-minded people reinforce each other's opinion, but do not get exposed to the views of the opposing side. In this paper we study algorithmic techniques for bridging these chambers, and thus reduce controversy. Specifically, we represent the discussion on a controversial issue with an endorsement graph, and cast our problem as an edge-recommendation problem on this graph. The goal of the recommendation is to reduce the controversy score of the graph, which is measured by a recently-developed metric based on random walks. At the same time, we take into account the acceptance probability of the recommended edge, which represents how likely the edge is to materialize in the endorsement graph.

Data generated on location-based social networks provide rich information on the whereabouts of urban dwellers. Specifically, such data reveal who spends time where, when, and on what type of activity (e.g., shopping at a mall, or dining at a restaurant). That information can, in turn, be used to describe city regions in terms of activity that takes place therein. For example, the data might reveal that citizens visit one region mainly for shopping in the morning, while another for dining in the evening. Furthermore, once such a description is available, one can ask more elaborate questions. For example, one might ask what features distinguish one region from another - some regions might be different in terms of the type of venues they host and others in terms of the visitors they attract. As another example, one might ask which regions are similar across cities. In this paper, we present a method to answer such questions using publicly shared Foursquare data. Our analysis makes use of a probabilistic model, the features of which include the exact location of activity, the users who participate in the activity, as well as the time of the day and day of week the activity takes place. Compared to previous approaches to similar tasks, our probabilistic modeling approach allows us to make minimal assumptions about the data - which relieves us from having to set arbitrary parameters in our analysis (e.g., regarding the granularity of discovered regions or the importance of different features). We demonstrate how the model learned with our method can be used to identify the most likely and distinctive features of a geographical area, quantify the importance features used in the model, and discover similar regions across different cities. Finally, we perform an empirical comparison with previous work and discuss insights obtained through our findings.

We study the problem of discrepancy maximization on graphs: given a set of nodes Q of an underlying graph G, we aim to identify a connected subgraph of G that contains many more nodes from Q than other nodes. This variant of the discrepancy maximization problem extends the well-known notion of “bump hunting” in the Euclidean space [1]. We consider the problem under two access models. In the unrestricted-access model, the whole graph G is given as input, while in the local-access model we can only retrieve the neighbors of a given node in G using a possibly slow and costly interface. We prove that the basic problem of discrepancy maximization on graphs is NP-hard, and empirically evaluate the performance of four heuristics for solving it. For the local-access model, we consider three different algorithms that aim to recover a part of G large enough to contain an optimal solution, while using only a small number of calls to the neighbor-function interface. We perform a thorough experimental evaluation in order to understand the trade offs between the proposed methods and their dependencies on characteristics of the input graph.

Which topics spark the most heated debates in social media? Identifying these topics is a first step towards creating systems which pierce echo chambers. In this paper, we perform a systematic methodological study of controversy detection using social media network structure and content.

We study the problem of discrepancy maximization on graphs: given a set of nodes Q of an underlying graph G, we aim to identify a connected subgraph of G that contains many more nodes from Q than other nodes. This variant of the discrepancy-maximization problem extends the well-known notion of “bump hunting” in the Euclidean space. We consider the problem under two access models. In the unrestricted-access model, the whole graph G is given as input, while in the local-access model we can only retrieve the neighbors of a given node in G using a possibly slow and costly interface. We prove that the basic problem of discrepancy maximization on graphs is NP-hard, and empirically evaluate the performance of four heuristics for solving it. For the local-access model we consider three different algorithms that aim to recover a part of G large enough to contain an optimal solution, while using only a small number of calls to the neighbor-function interface. We perform a thorough experimental evaluation in order to understand the trade offs between the proposed methods and their dependencies on characteristics of the input graph.

We study a new notion of graph centrality based on absorbing random walks. Given a graph G = (V, E) and a set of query nodes Q ⊆ V, we aim to identify the k most central nodes in G with respect to Q. Specifically, we consider central nodes to be absorbing for random walks that start at the query nodes Q. The goal is to find the set of k central nodes that minimizes the expected length of a random walk until absorption. The proposed measure, which we call k absorbing random-walk centrality, favors diverse sets, as it is beneficial to place the k absorbing nodes in different parts of the graph so as to “intercept” random walks that start from different query nodes. Although similar problem definitions have been considered in the literature, e.g., in information-retrieval settings where the goal is to diversify web-search results, in this paper we study the problem formally and prove some of its properties. We find that the problem is NP-hard, while the objective function is monotone and supermodular, implying that a greedy algorithm provides solutions with an approximation guarantee. On the other hand, the greedy algorithm involves expensive matrix operations that make it prohibitive to employ on large datasets. To confront this challenge, we explore the performance of efficient heuristics.

We present Spine, an efficient algorithm for finding the "backbone" of an influence network. Given a social graph and a log of past propagations, we build an instance of the independent-cascade model that describes the propagations. We aim at reducing the complexity of that model, while preserving most of its accuracy in describing the data.

We present TwitterMonitor, a system that performs trend detection over the Twitter stream. The system identifies emerging topics (i.e. 'trends') on Twitter in real time and provides meaningful analytics that synthesize an accurate description of each topic. Users interact with the system by ordering the identified trends using different criteria and submitting their own description for each trend.

Activity in social media such as blogs, micro-blogs, social networks, etc is manifested via interaction that involves text, images, links and other information items. Naturally, some items attract more attention than others, expressed with large volumes of linking, commenting or tagging activity, to name a few examples. Moreover, high attention can be indicative of emerging events, breaking news or generally indicate information items of interest to a vast set of people. The numbers associated with digital social activity are astonishing: in excess of millions of blog posts, tweets and forums updates per day, millions of tags in photos, news articles or blogs. Being able to identify information items that gather much attention in such a real time information collective is a challenging task.

User generated content that appears on weblogs, wikis and social networks has been increasing at an unprecedented rate. The wealth of information produced by individuals from different geographical locations presents a challenging task of intelligent processing. In this paper, we introduce a methodology to identify notable geographically focused events out of this collection of user generated information. At the heart of our proposal lie efficient algorithms that identify geographically focused information bursts, attribute them to demographic factors and identify sets of descriptive keywords. We present the results of a prototype evaluation of our algorithms on BlogScope, a large-scale social media warehousing platform. We demonstrate the scalability and practical utility of our proposal running on top of a multi-terabyte text collection.

Activity and user engagement in social media such as web logs, wikis, online forums or social networks has been increasing at unprecedented rates. In relation to social behavior in various human activities, user activity in social media indicates the existence of individuals that consistently drive or stimulate 'discussions' in the online world. Such individuals are considered as 'starters' of online discussions in contrast with 'followers' that primarily engage in discussions and follow them.

In this paper we perform an extensive theoretical and experimental study on common synopsis construction algorithms, with emphasis on wavelet based techniques, that take under consideration query workload statistics. Our goal is to compare, "expensive" quadratic time algorithms with "cheap" near-linear time algorithms, particularly when the latter are not optimal and/or not workload-aware for the problem at hand. Further, we present the first known algorithm for constructing wavelet synopses for a special class of range-sum query workloads. Our experimental results, clearly justify the necessity for designing workload-aware algorithms, especially in the case of range-sum queries.