Data describing networks—such as communication networks, transaction networks, disease transmission networks, collaboration networks, etc.—are becoming increasingly available. While observational data can be useful, it often only hints at the actual underlying process that governs interactions and attributes. For example, an email communication network provides insight into its users and their relationships, but is not the same as the “real” underlying social network. In this article, we introduce the problem of graph identification, i.e., discovering the latent graph structure underlying an observed network. We cast the problem as a probabilistic inference task, in which we must infer the nodes, edges, and node labels of a hidden graph, based on evidence. This entails solving several canonical problems in network analysis: entity resolution (determining when two observations correspond to the same entity), link prediction (inferring the existence of links), and node labeling (inferring hidden attributes). While each of these subproblems has been well studied in isolation, here we consider them as a single, collective task. We present a simple, yet novel, approach to address all three subproblems simultaneously. Our approach, which we refer to as C3, consists of a collection of Coupled Collective Classifiers that are applied iteratively to propagate inferred information among the subproblems. We consider variants of C3 using different learning and inference techniques and empirically demonstrate that C3 is superior, both in terms of predictive accuracy and running time, to state-of-the-art probabilistic approaches on four real problems.

Entity Resolution (ER) is the task of finding and merging entities within a single data source or across several data sources that represent the same real world entity. Evaluating ER results is a very important procedure used to ensure how accurate and correct an ER algorithm is. There have been several measures proposed and used for ER evaluation (e.g., pairwise F1, B3 F1, CEAF). With so many new and existing evaluation measures, there is a need to survey, characterize, and evaluate these measures. In this paper, we explore these ER evaluation measures. First, we provide a general definition to the ER problem. Next, we define many of the widely used measures, including a discussion of the different advantages and disadvantages of each measure. We also discuss trends in the use of the ER evaluation measures based across multiple domains. Finally, we provide examples of ER predictions and compute several of the measures we discuss to highlights differences and overlaps in the measures. In doing so, we hope to provide practitioners with a practical guide for understanding ER evaluation.

In this paper, on the one hand, we aim to give a review on literature dealing with the problem of supervised learning aided by additional unlabeled data. On the other hand, being a part of the author's first year PhD report, the paper serves as a frame to bundle related work by the author as well as numerous suggestions for potential future work. Therefore, this work contains more speculative and partly subjective material than the reader might expect from a literature review. We give a rigorous definition of the problem and relate it to supervised and unsupervised learning. The crucial role of prior knowledge is put forward, and we discuss the important notion of input-dependent regularization. We postulate a number of baseline methods, being algorithms or algorithmic schemes which can more or less straightforwardly be applied to the problem, without the need for genuinely new concepts. However, some of them might serve as basis for a genuine method. In the literature review, we try to cover the wide variety of (recent) work and to classify this work into meaningful categories. We also mention work done on related problems and suggest some ideas towards synthesis. Finally, we discuss some caveats and tradeoffs of central importance to the problem.

Introduction Whenever one asserts a claim of any kind, one engages in a commitment not just to that claim itself, but to a variety of other claims that follow in its wake, claims that, as we tend to say, follow logically from the original claim. To say that Smith and Jones are both great basketball players is to say something from which it follows that Smith is a great basketball player, that someone is a great basketball player, that there is something at which Smith is great, and so on. This general fact, that certain claims follow logically from others, is the central concern of a theory of logical consequence. Logical consequence is just the relation that connects a given claim or set of claims with those things that follow logically from it: to say that B is a logical consequence of A is simply to say that B follows logically from A. All of our ordinary reasoning turns on the recognition of this relation. When we notice, for example, that a certain prediction follows from a given theory, that a particular view is a consequence of some initial commitments, that a collection of premises entails a given conclusion, and so on, we are engaged in reasoning about logical consequence. The other logical properties and relations whose recognition is central to ordinary reasoning are closely related to, and can be defined in terms of, logical consequence. We say that an argument is valid iff its conclusion is a logical consequence of its premises; a set of claims G entails a claim a iff a is a logical consequence of G; a set of claims is consistent iff no contradiction is a logical consequence of it, and a claim a is independent of a set of claims G iff a is not a logical consequence of G. Finally, a claim is a logical truth iff it is a logical consequence of the empty set of claims. The investigation of logical consequence and related notions consists largely in the attempt (a) to give a systematic treatment of the extension of this relation, i.e. of the issue of which claims do in fact follow logically from which 2 others; and (b) to give an informative account of the nature of the relation. There is much room for debate about both of these issues. Though there is no doubt about the fact that some …

There is a growing interest in methods for analyzing data describing networks of all types, including information, biological, physical, and social networks. Typically the data describing these networks is observational, and thus noisy and incomplete; it is often at the wrong level of fidelity and abstraction for meaningful data analysis. This has resulted in a growing body of work on extracting, cleaning, and annotating network data. Unfortunately, much of this work is ad hoc and domain-specific. In this paper, we present the architecture of a data management system that enables efficient, declarative analysis of large-scale information networks. We identify a set of primitives to support the extraction and inference of a network from observational data, and describe a framework that enables a network analyst to easily implement and combine new extraction and analysis techniques, and efficiently apply them to large observation networks. The key insight behind our approach is to decouple, to the extent possible, (a) the operations that require traversing the graph structure (typically the computationally expensive step), from (b) the operations that do the modification and update of the extracted network. We present an analysis language based on Datalog, and show how to use it to cleanly achieve such decoupling. We briefly describe our prototype system that supports these abstractions. We include a preliminary performance evaluation of the system and show that our approach scales well and can efficiently handle a wide spectrum of data cleaning operations on network data.