Influence Maximization with Fairness at Scale

Machine Learning (ML) models trained on biased data can reproduce and even amplify these biases. Since such models are deployed to make decisions that can affect people's lives, ensuring their fairness is critical. One approach to mitigate possible unfairness of ML models is to map the input data into a less-biased new space by means of training the model on fair representations. Several methods based on adversarial learning have been proposed to learn fair representation by fooling an adversary in predicting the sensitive attribute (e.g., gender or race). However, adversarial-based learning can be too difficult to optimize in practice; besides, it penalizes the utility of the representation. Hence, in this research effort we train bias-free representations from the input data by inducing a uniform distribution over the sensitive attributes in the latent space. In particular, we propose a probabilistic framework that learns these representations by enforcing the correct reconstruction of the original data, plus the prediction of the attributes of interest while eliminating the possibility of predicting the sensitive ones. Our method leverages the inability of Deep Neural Networks (DNNs) to generalize when trained on a noisy label space to regularize the latent space. We use a network head that predicts a noisy version of the sensitive attributes in order to increase the uncertainty of their predictions at test time. Our experiments in two datasets demonstrated that the proposed model significantly improves fairness while maintaining the prediction accuracy of downstream tasks.

We address the problem of influence maximization when the social network is accompanied by diffusion cascades. In prior works, such information is used to compute influence probabilities, which is utilized by stochastic diffusion models in influence maximization. Motivated by the recent criticism on the effectiveness of diffusion models as well as the galloping advancements in influence learning, we propose IMINFECTOR (Influence Maximization with INFluencer vECTORs), a unified approach that uses representations learned from diffusion cascades to perform model-independent influence maximization that scales in real-world datasets. The first part of our methodology is a multi-task neural network that learns embeddings of nodes that initiate cascades (influencer vectors) and embeddings of nodes that participate in them (susceptible vectors). The norm of an influencer vector captures the ability of the node to create lengthy cascades and is used to estimate the expected influence spread and reduce the number of candidate seeds. In addition, the combination of influencer and susceptible vectors form the diffusion probabilities between nodes. These are used to reformulate the network as a bipartite graph and propose a greedy solution to influence maximization that retains the theoretical guarantees.We a pply our method in three sizable networks with diffusion cascades and evaluate it using cascades from future time steps. IMINFECTOR is able to scale in all of them and outperforms various competitive algorithms and metrics from the diverse landscape of influence maximization in terms of efficiency and seed set quality.

Fairness in machine learning refers to the development of models that do not systematically disadvantage individuals or groups based on sensitive attributes. One commonly adopted principle is fairness through unawareness, which holds that models should not explicitly incorporate sensitive attributes into the decision-making process. However, in some scenarios, such as healthcare, variables like sex and age are essential for accurate diagnoses and prognoses. Previous studies have assessed the influence of including or excluding such information during model training. Nonetheless, they have not considered Adversarial Debiasing, a classification algorithm specifically designed to promote equitable results. To address this gap, we propose a comprehensive empirical analysis to investigate the role of sensitive attributes in this algorithm. We experimentally evaluated Adversarial Debiasing across 20 settings using 23 datasets from varying domains, one predictive performance metric, three group fairness metrics, and a non-parametric statistical test. Our findings indicate that classifiers trained without including sensitive information in the input feature set produce more precise and fairer outcomes.

The popularity of social networks has grown manifolds in recent years because of various activities like fast propagation of ideas, publicity, and news. Influence maximization (IM) is one of the most highly studied problems in the field of social network analysis due to its business values. Influence maximization aims to identify influential nodes that can spread the information to the maximum number of nodes in the network through diffusion cascade. Traditional methods for IM include centrality based and greedy based measures. However, each method has some limitations. Recently, some methods of IM are introduced, which consider the presence of community structure in networks. Community structure has a significant impact on information diffusion, because of dense connections between the nodes in the community. In this paper, we propose a novel influence maximization algorithm using node seeding, label propagation, and community detection. We first use extended h-index centrality to detect the seed nodes and then use the label propagation technique to detect communities. Further, we merge smaller and related communities to a larger community with the help of a relationship matrix. Finally, top-k influential nodes from these communities are identified. These ideas lead to our proposed algorithm: Influence Maximization using Extended h-index, and Label Propagation with Relationship matrix (IM-ELPR). We adopt Independent Cascade (IC) information diffusion model to spread the information originating from chosen influential nodes. The proposed algorithm intends to identify influential nodes from different communities globally and does not depend on the specific community’s antecedent structural information. Experimental results performed on several real-life data sets reveal that the proposed algorithm performs better than many other existing popular algorithms.

Zuckerli is a scalable compression system meant for large real-world graphs. Graphs are notoriously challenging structures to store efficiently due to their linked nature, which makes it hard to separate them into smaller, compact components. Therefore, effective compression is crucial when dealing with large graphs, which can have billions of nodes and edges. Furthermore, a good compression system should give the user fast and reasonably flexible access to parts of the compressed data without requiring full decompression, which may be unfeasible on their system. Compared to WebGraph, the de-facto standard in compressing real-world graphs, Zuckerli improves multiple aspects by using advanced compression techniques and novel heuristic graph algorithms. Zuckerli can produce both a compressed representation for storage and one which allows fast direct access to the adjacency lists of the compressed graph without decompressing the entire graph. We validate its effectiveness on real-world graphs with up to a billion nodes and 90 billion edges, conducting an extensive experimental evaluation of both compression density and decompression performance. We show that Zuckerli-compressed graphs are 10% to 29% smaller, and more than 20% in most cases, with a resource usage for decompression comparable to that of WebGraph.

The new generation of social networks contains billions of nodes and edges. Managing and mining this data is a new academic and industrial challenge. Influence maximization is the problem of finding a set of nodes in a social network that result in the highest amount of influence diffusion. Many research works have been developed, which focus exclusively on the efficiency of algorithms, but overlook some features of social network data such as time sensitivity and the practicality in a large scale. Furthermore, the new era of 'big data' is changing dramatically right before our eyes - the increase of big data growth gives all researchers many challenges as well as opportunities. This paper proposes two new models TIC and TLT and considers the time constraint during the influence spreading process in practice. Empirical studies on different synthetic and real large scale social networks demonstrate that our models together with solutions on both Hadoop and Spark platforms are more practical as well as providing a regulatory mechanism for enhancing influence maximization. Not only that but also outperforming most existing alternative algorithms.

Graphs appear in numerous applications including cyber security, the Internet, social networks, protein networks, recommendation systems, citation networks, and many more. Graphs with millions or even billions of nodes and edges are common-place. How to store such large graphs efficiently? What are the core operations/queries on those graph? How to answer the graph queries quickly? We propose Gbase, an efficient analysis platform for large graphs. The key novelties lie in (1) our storage and compression scheme for a parallel, distributed settings and (2) the carefully chosen graph operations and their efficient implementations. We designed and implemented an instance of Gbase using Mapreduce/Hadoop. Gbase provides a parallel indexing mechanism for graph operations that both saves storage space, as well as accelerates query responses. We run numerous experiments on real and synthetic graphs, spanning billions of nodes and edges, and we show that our proposed Gbase is indeed fast, scalable, and nimble, with significant savings in space and time.

An important motivation for research in location privacy has been to protect against user profiling, i.e., inferring a user’s political affiliation, wealth level, sexual preferences, religious beliefs, and other sensitive attributes. Existing approaches focus on distorting or suppressing individual locations, but we argue that, for directly protecting against profiling, it is more appropriate to focus on the frequency with which various locations are visited – in other words, the histogram of a user’s locations. We introduce and explore a new privacy notion, namely, on-the-fly privacy for location histograms, in which a mobile user repeatedly submits obfuscated locations to a Location-Based Service aiming for the resulting histogram to resemble a target profile or differ from it. For example, she may want to avoid looking wealthy or to resemble a health-conscious person. We describe how to design concrete privacy mechanisms that operate under different assumptions on, e.g., the user’s mobility, including provably optimal mechanisms. We use a mobility dataset with 1083 users to illustrate how these mechanisms achieve privacy while minimizing the quality loss caused by the location obfuscation, in the context of two types of Location-Based Services: nearest-PoI, and geofence.

Data driven-based approaches including deep neural networks (DNN) have shown promise in various fields. Such techniques tend to require significant training for good convergence. Model-based approaches, however, provide practical data efficient solutions often with insightful and intuitive interpretations. A hybrid approach that employs data driven techniques aided by knowledge from model-based approaches may help reduce required training and improve convergence rates. This work investigates the potential of deep learning techniques to detect radar targets while accelerating the learning process via use of expert/domain knowledge from model-based algorithms for channel matrix-based cognitive sonar/radar. The channel matrices characterize responses from target and clutter/reverberation. The architecture of the proposed DNN exploits the insights from the model-based generalized likelihood ratio test (GLRT) statistic presented in our previous work, and hence, the resulting DNN algorithm benefits from the merits of both the model-based and data-driven approaches. Our proposed DNN architecture utilizes the secondary data for clutter channel estimation via the maximum-likelihood approach, and thus, requires little to no retraining with the changing clutter environment. We compare the detection performance of model-aided deep learning-based algorithms with that of traditional model-based techniques and pure data-driven DNN approaches using receiver operating characteristic (ROC) curves from Monte Carlo simulations.

Online social networks (OSNs) have witnessed their prosperous developments in recent years. Based on the established relations among individuals in OSNs, ideas and opinions can be spread via a word-of-mouth effect. To exploit this effect, many companies have taken OSNs as the major advertising channels to promote their products, which has motivated substantial research on viral marketing. Among them, influence maximization, seed minimization, and profit maximization have been extensively studied in the past decades. These three problems are mainly studied in the non-adaptive setting in the literature, i.e., seed user selection phase does not involve any observations on influence propagation in reality. In this thesis, we extend profit maximization into a generalized version as target profit maximization and investigate these three problems in the adaptive setting, formulated as adaptive influence maximization, adaptive seed minimization, and adaptive target profit maximization respectively. Firstly, we study the adaptive influence maximization (IM) problem. Given a social network G and an integer k, the influence maximization problem asks for a seed set S of k nodes from G to maximize the expected number of nodes activated via an influence propagation. In the adaptive setting, the k seed nodes are selected in batches of equal size b, such that the i-th batch is identified after the actual influence results of the former i-1 batches are observed. We propose the first practical algorithm for the adaptive IM problem that could provide the worst-case approximation guarantee of 1-e^α(ε-1) with high probability, where ε∊(0, 1) is a user-specified parameter and α= 1-(1-1/b)^b is set by the batch size b. Our approach is based on a novel general framework AdaptGreedy which could be instantiated by any existing non-adaptive IM algorithms with expected approximation guarantee. In this regard, we design a novel non-adaptive IM algorithm called EPIC which not only provides strong expected approximation guarantee, but also presents superior performance compared with the existing IM algorithms. Meanwhile, we clarify some existing misunderstandings in recent work and shed light on further study of the adaptive IM problem. Subsequently, we consider the adaptive seed minimization (SM) problem. As a dual problem of influence maximization, the seed minimization problem asks for the minimum number of seed nodes to influence a required number η of users in a given social network G. In the adaptive setting, the seed nodes are selected in several batches, such that the choice of current batch may exploit information about the actual influence of the previous batches. We devise a novel algorithm, ASTI, which addresses the adaptive seed minimization problem in O (η(m+n)/ε^2 ln n) expected time and offers an approximation guarantee of (ln η+1)^2/((1-(1-1/b)^b)(1-1/e)(1-ε)) in expectation, where η is the targeted number of influenced nodes, b is size of each seed node batch, and ε∊(0, 1) is a user-specified parameter. To the best of our knowledge, ASTI is the first algorithm that provides such an approximation guarantee without incurring prohibitive computation overhead. Eventually, we explore the adaptive target profit maximization (TPM) problem. Given a social network G, the profit maximization (PM) problem asks for a set of seed nodes to maximize the profit, i.e., revenue of influence spread less the cost of seed selection. The TPM problem, which generalizes the PM problem, aims to select a subset of seed nodes from a given target user set T to maximize the profit. In the adaptive setting, the seed users are selected through multiple batches, such that the selection of current batch exploits the knowledge of actual influence in the previous batches. To acquire an overall understanding, we study adaptive TPM under the oracle model and the noise model, to which we propose ADG and AddATP algorithms with strong theoretical guarantees respectively. In addition, we propose the idea of hybrid error to better handle the sampling errors under the noise model and design a novel algorithm HATP that boosts the efficiency of AddATP significantly.

Influence maximization is one of the most fundamental problems in social network analysis due to its significant impact on viral marketing and targeted advertisements. Considerable amount of works has extensively studied to analyze the influence in social networks, but most of the existing works unfortunately neglected the fact that the location information in the cyber-physical world could also play an important role in the influence prorogation. Furthermore, even though a few works consider a little location information to enhance the influence maximization, they do not discuss any privacy issue of the model and expose user's location information directly to the public. This paper considers the problem of influence maximization in both GPS-enabled cyber-physical and online social networks with privacy reservation. We propose one novel model merging both GPS data of cyber-physical network and relationship data of online social network together in a unified framework, then we provide an efficient algorithm to solve the influence maximization problem in the framework. Besides the influence maximization problem, our framework could also support other applications involving both cyber-physical and online social networks. Further more, to protect the sensitive location and link information, we also provide corresponding techniques to protect the privacy during the whole influence maximization process. Empirical studies of real life datasets demonstrate the power of engaging location information to influence maximization, and suggest that our resolution outperforming most existing alternative algorithms with the protection of location privacy.

In recent years, due to the surge in popularity of social-networking web sites, considerable interest has arisen regarding influence maximization in social networks. Given a social network structure, the problem of influence maximization is to determine a minimum set of nodes that could maximize the spread of influences. With a large-scale social network, the efficiency and practicability of such algorithms are critical. Although many recent studies have focused on the problem of influence maximization, these works in general are time-consuming when a social network is large-scale. In this paper, we propose two novel algorithms, CDH-Kcut and Community and Degree Heuristic on Kcut/SHRINK, to solve the influence maximization problem based on a realistic model. The algorithms utilize the community structure, which significantly decreases the number of candidates of influential nodes, to avoid information overlap. The experimental results on both synthetic and real datasets indicate that our algorithms not only significantly outperform the state-of-the-art algorithms in efficiency but also possess graceful scalability.

Distributed-memory clusters are used for in-memory processing of very large graphs with billions of nodes and edges. This requires partitioning the graph among the machines in the cluster. When a graph is partitioned, a node in the graph may be replicated on several machines, and communication is required to keep these replicas synchronized. Good partitioning policies attempt to reduce this synchronization overhead while keeping the computational load balanced across machines. A number of recent studies have looked at ways to control replication of nodes, but these studies are not conclusive because they were performed on small clusters with eight to sixteen machines, did not consider work-efficient data-driven algorithms, or did not optimize communication for the partitioning strategies they studied. This paper presents an experimental study of partitioning strategies for work-efficient graph analytics applications on large KNL and Skylake clusters with up to 256 machines using the Gluon communication runtime which implements partitioning-specific communication optimizations. Evaluation results show that although simple partitioning strategies like Edge-Cuts perform well on a small number of machines, an alternative partitioning strategy called Cartesian Vertex-Cut (CVC) performs better at scale even though paradoxically it has a higher replication factor and performs more communication than Edge-Cut partitioning does. Results from communication micro-benchmarks resolve this paradox by showing that communication overhead depends not only on communication volume but also on the communication pattern among the partitions. These experiments suggest that high-performance graph analytics systems should support multiple partitioning strategies, like Gluon does, as no single graph partitioning strategy is best for all cluster sizes. For such systems, a decision tree for selecting a good partitioning strategy based on characteristics of the computation and the cluster is presented.

We have entered the age of big data. Massive datasets are now common in science, government and enterprises. Yet, making sense of these data remains a fundamental challenge. Where do we start our analysis? Where to go next? How to visualize our findings? We answers these questions by bridging Data Mining and Human- Computer Interaction (HCI) to create tools for making sense of graphs with billions of nodes and edges, focusing on: (1) Attention Routing: we introduce this idea, based on anomaly detection, that automatically draws people’s attention to interesting areas of the graph to start their analyses. We present three examples: Polonium unearths malware from 37 billion machine-file relationships; NetProbe fingers bad guys who commit auction fraud. (2) Mixed-Initiative Sensemaking: we present two examples that combine machine inference and visualization to help users locate next areas of interest: Apolo guides users to explore large graphs by learning from few examples of user interest; Graphite finds interesting subgraphs, based on only fuzzy descriptions drawn graphically. (3) Scaling Up: we show how to enable interactive analytics of large graphs by leveraging Hadoop, staging of operations, and approximate computation. This thesis contributes to data mining, HCI, and importantly their intersection, including: interactive systems and algorithms that scale; theories that unify graph mining approaches; and paradigms that overcome fundamental challenges in visual analytics. Our work is making impact to academia and society: Polonium protects 120 million people worldwide from malware; NetProbe made headlines on CNN, WSJ and USA Today; Pegasus won an opensource software award; Apolo helps DARPA detect insider threats and prevent exfiltration. We hope our Big Data Mantra “Machine for Attention Routing, Human for Interaction” will inspire more innovations at the crossroad of data mining and HCI.

Given a large graph with millions or billions of nodes and edges, like a who-follows-whom Twitter graph, how do we scalably compute its statistics, summarize its patterns, spot anomalies, visualize and make sense of it? We present OPAvion, a graph mining system that provides a scalable, interactive workflow to accomplish these analysis tasks. OPAvion consists of three modules: (1) The Summarization module (Pegasus) operates off-line on massive, disk-resident graphs and computes graph statistics, like PageRank scores, connected components, degree distribution, triangles, etc.; (2) The Anomaly Detection module (OddBall) uses graph statistics to mine patterns and spot anomalies, such as nodes with many contacts but few interactions with them (possibly telemarketers); (3) The Interactive Visualization module (Apolo) lets users incrementally explore the graph, starting with their chosen nodes or the flagged anomalous nodes; then users can expand to the nodes' vicinities, label them into categories, and thus interactively navigate the interesting parts of the graph.In our demonstration, we invite our audience to interact with OPAvion and try out its core capabilities on the Stack Overflow Q&A graph that describes over 6 million questions and answers among 650K users.