Ralitsa Angelova

Towards time-aware link prediction in evolving social networks

Prediction of links - both new as well as recurring - in a social network representing interactions between individuals is an important problem. In the recent years, there is significant interest in methods that use only the graph structure to make predictions. However, most of them consider a single snapshot of the network as the input, neglecting an important aspect of these social networks viz., their evolution over time. In this work, we investigate the value of incorporating the history information available on the interactions (or links) of the current social network state. Our results unequivocally show that time-stamps of past interactions significantly improve the prediction accuracy of new and recurrent links over rather sophisticated methods proposed recently. Furthermore, we introduce a novel testing method which reflects the application of link prediction better than previous approaches.

Graph-based text classification: learn from your neighbors

Automatic classification of data items, based on training samples, can be boosted by considering the neighborhood of data items in a graph structure (e.g., neighboring documents in a hyperlink environment or co-authors and their publications for bibliographic data entries). This paper presents a new method for graph-based classification, with particular emphasis on hyperlinked text documents but broader applicability. Our approach is based on iterative relaxation labeling and can be combined with either Bayesian or SVM classifiers on the feature spaces of the given data items. The graph neighborhood is taken into consideration to exploit locality patterns while at the same time avoiding overfitting. In contrast to prior work along these lines, our approach employs a number of novel techniques: dynamically inferring the link/class pattern in the graph in the run of the iterative relaxation labeling, judicious pruning of edges from the neighborhood graph based on node dissimilarities and node degrees, weighting the influence of edges based on a distance metric between the classification labels of interest and weighting edges by content similarity measures. Our techniques considerably improve the robustness and accuracy of the classification outcome, as shown in systematic experimental comparisons with previously published methods on three different real-world datasets.

A neighborhood-based approach for clustering of linked document collections

This paper addresses the problem of automatically structuring linked document collections by using clustering. In contrast to traditional clustering, we study the clustering problem in the light of available link structure information for the data set (e.g., hyperlinks among web documents or co-authorship among bibliographic data entries). Our approach is based on iterative relaxation of cluster assignments, and can be built on top of any clustering algorithm. This technique results in higher cluster purity, better overall accuracy, and make self-organization more robust.

Graffiti: graph-based classification in heterogeneous networks

We address the problem of multi-label classification in heterogeneous graphs, where nodes belong to different types and different types have different sets of classification labels. We present a novel approach that aims to classify nodes based on their neighborhoods. We model the mutual influence of nodes as a random walk in which the random surfer aims at distributing class labels to nodes while walking through the graph. When viewing class labels as “colors”, the random surfer is essentially spraying different node types with different color palettes; hence the name Graffiti of our method. In contrast to previous work on topic-based random surfer models, our approach captures and exploits the mutual influence of nodes of the same type based on their connections to nodes of other types. We show important properties of our algorithm such as convergence and scalability. We also confirm the practical viability of Graffiti by an experimental study on subsets of the popular social networks Flickr and LibraryThing. We demonstrate the superiority of our approach by comparing it to three other state-of-the-art techniques for graph-based classification.

In Defense of Max Planck

The Max Planck PhDnet, representing about 4000 Max Planck graduate students, takes issue with the unfounded claim by Widmar Tanner that a disconnect between German universities and Max Planck Institutes (MPIs) leads to MPI graduates that are “at best average” (“Max Planck accused of hobbling