Tina Eliassi-Rad

Collective Classification in Network Data

Many real-world applications produce networked data such as the world-wide web (hypertext documents connected via hyperlinks), social networks (for example, people connected by friendship links), communication networks (computers connected via communication links) and biological networks (for example, protein interaction networks). A recent focus in machine learning research has been to extend traditional machine learning classification techniques to classify nodes in such networks. In this article, we provide a brief introduction to this area of research and how it has progressed during the past decade. We introduce four of the most widely used inference algorithms for classifying networked data and empirically compare them on both synthetic and real-world data.

Visual Analysis of Large Heterogeneous Social Networks by Semantic and Structural Abstraction

Social network analysis is an active area of study beyond sociology. It uncovers the invisible relationships between actors in a network and provides understanding of social processes and behaviors. It has become an important technique in a variety of application areas such as the Web, organizational studies, and homeland security. This paper presents a visual analytics tool, OntoVis, for understanding large, heterogeneous social networks, in which nodes and links could represent different concepts and relations, respectively. These concepts and relations are related through an ontology (also known as a schema). OntoVis is named such because it uses information in the ontology associated with a social network to semantically prune a large, heterogeneous network. In addition to semantic abstraction, OntoVis also allows users to do structural abstraction and importance filtering to make large networks manageable and to facilitate analytic reasoning. All these unique capabilities of OntoVis are illustrated with several case studies

Gelling, and melting, large graphs by edge manipulation

Controlling the dissemination of an entity (e.g., meme, virus, etc) on a large graph is an interesting problem in many disciplines. Examples include epidemiology, computer security, marketing, etc. So far, previous studies have mostly focused on removing or inoculating nodes to achieve the desired outcome. We shift the problem to the level of edges and ask: which edges should we add or delete in order to speed-up or contain a dissemination? First, we propose effective and scalable algorithms to solve these dissemination problems. Second, we conduct a theoretical study of the two problems and our methods, including the hardness of the problem, the accuracy and complexity of our methods, and the equivalence between the different strategies and problems. Third and lastly, we conduct experiments on real topologies of varying sizes to demonstrate the effectiveness and scalability of our approaches.

On the Vulnerability of Large Graphs

Given a large graph, like a computer network, which k nodes should we immunize (or monitor, or remove), to make it as robust as possible against a computer virus attack? We need (a) a measure of the ‘Vulnerability’ of a given network, b) a measure of the ‘Shield-value’ of a specific set of k nodes and (c) a fast algorithm to choose the best such k nodes. We answer all these three questions: we give the justification behind our choices, we show that they agree with intuition as well as recent results in immunology. Moreover, we propose Net Shield, a fast and scalable algorithm. Finally, we give experiments on large real graphs, where Net Shield achieves tremendous speed savings exceeding 7 orders of magnitude, against straightforward competitors.

Applying latent dirichlet allocation to group discovery in large graphs

This paper introduces LDA-G, a scalable Bayesian approach to finding latent group structures in large real-world graph data. Existing Bayesian approaches for group discovery (such as Infinite Relational Models) have only been applied to small graphs with a couple of hundred nodes. LDA-G (short for Latent Dirichlet Allocation for Graphs) utilizes a well-known topic modeling algorithm to find latent group structure. Specifically, we modify Latent Dirichlet Allocation (LDA) to operate on graph data instead of text corpora. Our modifications reflect the differences between real-world graph data and text corpora (e.g., a nodes neighbor count vs. a documents word count). In our empirical study, we apply LDA-G to several large graphs (with thousands of nodes) from PubMed (a scientific publication repository). We compare LDA-Gs quantitative performance on link prediction with two existing approaches: one Bayesian (namely, Infinite Relational Model) and one non-Bayesian (namely, Cross-association). On average, LDA-G outperforms IRM by 15% and Cross-association by 25% (in terms of area under the ROC curve). Furthermore, we demonstrate that LDA-G can discover useful qualitative information.

A Probabilistic Model for Using Social Networks in Personalized Item Recommendation

Preference-based recommendation systems have transformed how we consume media. By analyzing usage data, these methods uncover our latent preferences for items (such as articles or movies) and form recommendations based on the behavior of others with similar tastes. But traditional preference-based recommendations do not account for the social aspect of consumption, where a trusted friend might point us to an interesting item that does not match our typical preferences. In this work, we aim to bridge the gap between preference- and social-based recommendations. We develop social Poisson factorization (SPF), a probabilistic model that incorporates social network information into a traditional factorization method; SPF introduces the social aspect to algorithmic recommendation. We develop a scalable algorithm for analyzing data with SPF, and demonstrate that it outperforms competing methods on six real-world datasets; data sources include a social reader and Etsy.

APATE: A novel approach for automated credit card transaction fraud detection using network-based extensions

In the last decade, the ease of online payment has opened up many new opportunities for e-commerce, lowering the geographical boundaries for retail. While e-commerce is still gaining popularity, it is also the playground of fraudsters who try to misuse the transparency of online purchases and the transfer of credit card records. This paper proposes APATE, a novel approach to detect fraudulent credit card transactions conducted in online stores. Our approach combines (1) intrinsic features derived from the characteristics of incoming transactions and the customer spending history using the fundamentals of RFM (Recency - Frequency - Monetary); and (2) network-based features by exploiting the network of credit card holders and merchants and deriving a time-dependent suspiciousness score for each network object. Our results show that both intrinsic and network-based features are two strongly intertwined sides of the same picture. The combination of these two types of features leads to the best performing models which reach AUC-scores higher than 0.98.

A System for Building Intelligent Agents that Learn to Retrieve and Extract Information

We present a system for rapidly and easily building instructable and self-adaptive software agents that retrieve and extract information. Our Wisconsin Adaptive Web Assistant (WAWA) constructs intelligent agents by accepting user preferences in the form of instructions. These user-provided instructions are compiled into neural networks that are responsible for the adaptive capabilities of an intelligent agent. The agent’s neural networks are modified via user-provided and system-constructed training examples. Users can create training examples by rating Web pages (or documents), but more importantly WAWA’s agents uses techniques from reinforcement learning to internally create their own examples. Users can also provide additional instruction throughout the life of an agent. Our experimental evaluations on a ‘home-page finder’ agent and a ‘seminar-announcement extractor’ agent illustrate the value of using instructable and adaptive agents for retrieving and extracting information.

Leveraging label-independent features for classification in sparsely labeled networks: an empirical study

We address the problem of within-network classification in sparsely labeled networks. Recent work has demonstrated success with statistical relational learning (SRL) and semi-supervised learning (SSL) on such problems. However, both approaches rely on the availability of labeled nodes to infer the values of missing labels. When few labels are available, the performance of these approaches can degrade. In addition, many such approaches are sensitive to the specific set of nodes labeled. So, although average performance may be acceptable, the performance on a specific task may not. We explore a complimentary approach to within-network classification, based on the use of label-independent (LI) features - i.e., features calculated without using the values of class labels. While previous work has made some use of LI features, the effects of these features on classification performance have not been extensively studied. Here, we present an empirical study in order to better understand these effects. Through experiments on several real-world data sets, we show that the use of LI features produces classifiers that are less sensitive to specific label assignments and can lead to performance improvements of over 40% for both SRL- and SSL-based classifiers. We also examine the relative utility of individual LI features; and show that, in many cases, it is a combination of a few diverse network-based structural characteristics that is most informative.

Two heads better than one: pattern discovery in time-evolving multi-aspect data

Data stream values are often associated with multiple aspects. For example each value observed at a given time-stamp from environmental sensors may have an associated type (e.g., temperature, humidity, etc.) as well as location. Time-stamp, type and location are the three aspects, which can be modeled using a tensor (high-order array). However, the time aspect is special, with a natural ordering, and with successive time-ticks having usually correlated values. Standard multiway analysis ignores this structure. To capture it, we propose 2 Heads Tensor Analysis (2-heads), which provides a qualitatively different treatment on time. Unlike most existing approaches that use a PCA-like summarization scheme for all aspects, 2-heads treats the time aspect carefully. 2-heads combines the power of classic multilinear analysis with wavelets, leading to a powerful mining tool. Furthermore, 2-heads has several other advantages as well: (a) it can be computed incrementally in a streaming fashion, (b) it has a provable error guarantee and, (c) it achieves significant compression ratio against competitors. Finally, we show experiments on real datasets, and we illustrate how 2-heads reveals interesting trends in the data. This is an extended abstract of an article published in the Data Mining and Knowledge Discovery journal.