Temporal-Relational Classifiers for Prediction in Evolving Domains

Abstract

Many relational domains contain temporal information and dynamics that are important to model (e.g., social networks, protein networks). However, past work in relational learning has focused primarily on modeling static "snapshots" of the data and has largely ignored the temporal dimension of these data. In this work, we extend relational techniques to temporally-evolving domains and outline a representational framework that is capable of modeling both temporal and relational dependencies in the data. We develop efficient learning and inference techniques within the framework by considering a restricted set of temporalrelational dependencies and using parameter-tying methods to generalize across relationships and entities. More specifically, we model dynamic relational data with a twophase process, first summarizing the temporal-relational information with kernel smoothing, and then moderating attribute dependencies with the summarized relational information. We develop a number of novel temporal-relational models using the framework and then show that the current approaches to modeling static relational data are special cases within the framework. We compare the new models to the competing static relational methods on three real-world datasets and show that the temporal-relational models consistently outperform the relational models that ignore temporal information-achieving significant reductions in error ranging from 15% to 70%.

Predicting group stability in online social networks

“…Since we have timestamped (year of publication) data detailing the activities (publications) of a person (author) in a community (conference/journal), we can define a measure that reflects all of the above properties by adapting the exponential summarization kernel described in [23]. Let N1, N2 .…”

Randomization tests for distinguishing social influence and homophily effects

“…Since autocorrelation is the primary motivation to use relational and network models over conventional machine learning techniques, it stands to reason that a better understanding of the causes of autocorrelation will inform the development of improved models and learning algorithms. For example, although previous work in relational learning and statistical network analysis has focused primarily on static graphs, recent efforts have turned to the analysis of dynamic networks and development of temporally-evolving models (e.g., [9,21]). In order to deal with the enormous increase in dimensionality associated with modeling both temporal and relational dependencies, these methods restrict the set of dependencies that they consider (e.g., through choice of model form).…”

Time-Evolving Relational Classification and Ensemble Methods

“…Traditionally, relational classifiers have attempted to use all the data available in a network [18]. However, since the relevance of data may change over time (e.g., links become stale), learning the appropriate temporal granularity (i.e., range of timesteps) can improve classification accuracy.…”

“…Previous work in relational learning on attributed graphs either uses static network snapshots or significantly limits the amount of temporal information incorporated into the models. Sharan et al [18] assumes a strict representation that only uses kernel estimation for link weights, while GA-TVRC [9] uses a genetic algorithm to learn the link weights. SRPTs [11] incorporate temporal and spatial information in the relational attributes.…”