Abstract. The presence of autocorrelation provides strong motivation for using relational techniques for learning and inference. Autocorrelation is a statistical dependency between the values of the same variable on related entities and is a nearly ubiquitous characteristic of relational data sets. Recent research has explored the use of collective inference techniques to exploit this phenomenon. These techniques achieve significant performance gains by modeling observed correlations among class labels of related instances, but the models fail to capture a frequent cause of autocorrelation—the presence of underlying groups that influence the attributes on a set of entities. We propose a latent group model (LGM) for relational data, which discovers and exploits the hidden structures responsible for the observed autocorrelation among class labels. Modeling the latent group structure improves model performance, increases inference efficiency, and enhances our understanding of the datasets. We evaluate performance on three relational classification tasks and show that LGM outperforms models that ignore latent group structure when there is little known information with which to seed inference.

Intelligent agents must be able to handle the complexity and uncertainty of the real world. Logical AI has focused mainly on the former, and statistical AI on the latter. Markov logic combines the two by attaching weights to first-order formulas and viewing them as templates for features of Markov networks. Inference algorithms for Markov logic draw on ideas from satisfiability, Markov chain Monte Carlo and knowledge-based model construction. Learning algorithms are based on the voted perceptron, pseudo-likelihood and inductive logic programming. Markov logic has been successfully applied to problems in entity resolution, link prediction, information extraction and others, and is the basis of the open-source Alchemy system.

Many real-world applications of AI require both probability and first-order logic to deal with uncertainty and structural complexity. Logical AI has focused mainly on handling complexity, and statistical AI on handling uncertainty. Markov Logic Networks (MLNs) are a powerful representation that combine Markov Networks (MNs) and first-order logic by attaching weights to first-order formulas and viewing these as templates for features of MNs. State-of-theart structure learning algorithms of MLNs maximize the likelihood of a relational database by performing a greedy search in the space of candidates. This can lead to suboptimal results because of the incapability of these approaches to escape local optima. Moreover, due to the combinatorially explosive space of potential candidates these methods are computationally prohibitive. We propose a novel algorithm for learning MLNs structure, based on the Iterated Local Search (ILS) metaheuristic that explores the space of structures through a biased sampling of the set of local optima. The algorithm focuses the search not on the full space of solutions but on a smaller subspace defined by the solutions that are locally optimal for the optimization engine. We show through experiments in two real-world domains that the proposed approach improves accuracy and learning time over the existing state-of-the-art algorithms. 1

In collective classification, classes are predicted simultaneously for a group of related instances, rather than predicting a class for each instance separately. Collective classification has been widely used for classification on relational datasets. However, the inference procedure used in collective classification usually requires many iterations and thus is expensive. We propose stacked graphical learning, a meta-learning scheme in which a base learner is augmented by expanding one instance’s features with predictions on other related instances. Stacked graphical learning is efficient, especially during inference, capable of capturing dependencies easily, and can be implemented with any kind of base learner. In experiments on eight datasets, stacked graphical learning is 40 to 80 times faster than Gibbs sampling during inference. 1

Collective classification can significantly improve accuracy by exploiting relationships among instances. Although several collective inference procedures have been reported, they have not been thoroughly evaluated for their commonalities and differences. We introduce novel generalizations of three existing algorithms that allow such algorithmic and empirical comparisons. Our generalizations permit us to examine how cautiously or aggressively each algorithm exploits intermediate relational data, which can be noisy. We conjecture that cautious approaches that identify and preferentially exploit the more reliable intermediate data should outperform aggressive approaches. We explain why caution is useful and introduce three parameters to control the degree of caution. An empirical evaluation of collective classification algorithms, using two base classifiers on three data sets, supports our conjecture.

This paper surveys first-order probabilistic languages (FOPLs), which combine the expressive power of first-order logic with a probabilistic treatment of uncertainty. We provide a taxonomy that helps make sense of the profusion of FOPLs that have been proposed over the past fifteen years. We also emphasize the importance of representing uncertainty not just about the attributes and relations of a fixed set of objects, but also about what objects exist. This leads us to Bayesian logic, or BLOG, a new language for defining probabilistic models with unknown objects. We give a brief overview of BLOG syntax and semantics, and emphasize some of the design decisions that distinguish it from other languages. Finally, we consider the challenge of constructing FOPL models automatically from data.

There is extensive interest in mining data from full text. We have built a system called SLIF (for Subcellular Location Image Finder), which extracts information on one particular aspect of biology from a combination of text and images in journal articles. Associating the information from the text and image requires matching sub-figures with the sentences in the text. We introduce a stacked graphical model, a meta-learning scheme to augment a base learner by expanding features based on related instances, to match the labels of sub-figures with labels of sentences. The experimental results show a significant improvement in the matching accuracy of the stacked graphical model (81.3%) as compared with a relational dependency network (70.8%) or the current algorithm in SLIF (64.3%). 1.

Many data sets routinely captured by organizations are relational in nature - from marketing and sales transactions, to scientific observations and medical records. Relational data record characteristics of heterogeneous objects and persistent relationships among those objects (e.g., citation graphs, the World Wide Web, genomic structures). These data offer unique opportunities to improve model accuracy, and thereby decision-making, if machine learning techniques can effectively exploit the relational information. This work focuses on how to learn accurate statistical models of complex, relational data sets and develops two novel probabilistic models to represent, learn, and reason about statistical dependencies in these data. Relational dependency networks are the first relational model capable of learning general autocorrelation dependencies, an important class of statistical dependencies that are ubiquitous in relational data. Latent group models are the first relational model to generalize about the properties of underlying group structures to improve inference accuracy and efficiency. Not only do these two models offer performance gains over current relational models, but they also offer efficiency gains which will make relational modeling feasible for large, relational datasets where current methods are computationally intensive, if not intractable. We also formulate of a novel analysis framework to analyze relational model performance and ascribe errors to model learning and inference procedures. Within this framework, we explore the effects of data characteristics and representation choices on inference accuracy and investigate the mechanisms behind model performance. In particular, we show that the inference process in relational models can be a significant source of error and that relative model performance varies significantly across different types of relational data.

Abstract not found

Over the past few years, a number of approximate inference algorithms for networked data have been put forth. We empirically compare the performance of three of the popular algorithms: loopy belief propagation, mean field relaxation labeling and iterative classification. We rate each algorithm in terms of its robustness to noise, both in attribute values and correlations across links. A novel observation from our experiments is that loopy belief propagation faces difficulty when inferring over data with homophily, a common type of link correlation observed in relational data. 1.