Michael Salter-Townshend

Review of statistical network analysis: models, algorithms, and software

The analysis of network data is an area that is rapidly growing, both within and outside of the discipline of statistics. This review provides a concise summary of methods and models used in the statistical analysis of network data, including the Erdős–Renyi model, the exponential family class of network models, and recently developed latent variable models. Many of the methods and models are illustrated by application to the well-known Zachary karate dataset. Software routines available for implementing methods are emphasized throughout. The aim of this paper is to provide a review with enough detail about many common classes of network models to whet the appetite and to point the way to further reading.

Variational Bayesian inference for the Latent Position Cluster Model for network data

Analyzing Networks and Learning with Graphs Workshop at 23rd annual conference on Neural Information Processing Systems (NIPS 2009), Whister, December 11 2009

Role Analysis in Networks Using Mixtures of Exponential Random Graph Models

This article introduces a novel and flexible framework for investigating the roles of actors within a network. Particular interest is in roles as defined by local network connectivity patterns, identified using the ego-networks extracted from the network. A mixture of exponential-family random graph models (ERGM) is developed for these ego-networks to cluster the nodes into roles. We refer to this model as the ego-ERGM. An expectation-maximization algorithm is developed to infer the unobserved cluster assignments and to estimate the mixture model parameters using a maximum pseudo-likelihood approximation. We demonstrate the flexibility and utility of the method using examples of simulated and real networks.

The influence of network structures of Wikipedia discussion pages on the efficiency of WikiProjects

Abstract As a platform for discussion and communication, talk pages play an essential role in Wikipedia to facilitate coordination, sharing of information and knowledge resources among Wikipedians. In this work we explore the influence of network structures of these pages on the efficiency of WikiProjects. Project efficiency is measured as the amount of work done by project members in a quarter. The study uses the comments on WikiProject talk pages to construct communication networks. The structural properties of these networks are studied using ideas from social network theory. We develop three hypotheses about how network structures influence project effectiveness and examine the hypotheses using a longitudinal dataset of 362 WikiProjects. The evaluation suggests that an intermediate level of cohesion with a core of influential users dominating network flow improves effectiveness for a WikiProject, and that greater average membership tenure relates to project efficiency in a positive way. We discuss the implications of this analysis for the future management of WikiProjects.

Latent space models for multiview network data

Social relationships consist of interactions along multiple dimensions. In social networks, this means that individuals form multiple types of relationships with the same person (e.g., an individual will not trust all of his/her acquaintances). Statistical models for these data require understanding two related types of dependence structure: (i) structure within each relationship type, or network view, and (ii) the association between views. In this paper, we propose a statistical framework that parsimoniously represents dependence between relationship types while also maintaining enough flexibility to allow individuals to serve different roles in different relationship types. Our approach builds on work on latent space models for networks [see, e.g., J. Amer. Statist. Assoc.97 (2002) 1090-1098]. These models represent the propensity for two individuals to form edges as conditionally independent given the distance between the individuals in an unobserved social space. Our work departs from previous work in this area by representing dependence structure between network views through a multivariate Bernoulli likelihood, providing a representation of between-view association. This approach infers correlations between views not explained by the latent space model. Using our method, we explore 6 multiview network structures across 75 villages in rural southern Karnataka, India [Banerjee et al. (2013)].

Sentiment analysis of online media

A joint model for annotation bias and document classification is presented in the context of media sentiment analysis. We consider an Irish online media data set comprising online news articles with user annotations of negative, positive or irrelevant impact on the Irish economy. The joint model combines a statistical model for user annotation bias and a Naive Bayes model for the document terms. An EM algorithm is used to estimate the annotation bias model, the unobserved biases in the user annotations, the classifier parameters and the sentiment of the articles. The joint modeling of both the user biases and the classifier is demonstrated to be superior to estimation of the bias followed by the estimation of the classifier parameters.

Mixtures of biased sentiment analysers

Modelling bias is an important consideration when dealing with inexpert annotations. We are concerned with training a classifier to perform sentiment analysis on news media articles, some of which have been manually annotated by volunteers. The classifier is trained on the words in the articles and then applied to non-annotated articles. In previous work we found that a joint estimation of the annotator biases and the classifier parameters performed better than estimation of the biases followed by training of the classifier. An important question follows from this result: can the annotators be usefully clustered into either predetermined or data-driven clusters, based on their biases? If so, such a clustering could be used to select, drop or otherwise categorise the annotators in a crowdsourcing task. This paper presents work on fitting a finite mixture model to the annotators’ bias. We develop a model and an algorithm and demonstrate its properties on simulated data. We then demonstrate the clustering that exists in our motivating dataset, namely the analysis of potentially economically relevant news articles from Irish online news sources.

Fine-scale Inference of Ancestry Segments without Prior Knowledge of Admixing Groups

We present an algorithm for inferring ancestry segments and characterizing admixture events, which involve an arbitrary number of genetically differentiated groups coming together. This allows inference of the demographic history of the species, properties of admixing groups, identification of signatures of natural selection, and may aid disease gene mapping. The algorithm employs nested hidden Markov models to obtain local ancestry estimation along the genome for each admixed individual. In a range of simulations, the accuracy of these estimates equals or exceeds leading existing methods that return local ancestry. Moreover, and unlike these approaches, we do not require any prior knowledge of the relationship between sub-groups of donor reference haplotypes and the unseen mixing ancestral populations. Instead, our approach infers these in terms of conditional “copying probabilities”. In application to the Human Genome Diversity Panel we corroborate many previously inferred admixture events (e.g. an ancient admixture event in the Kalash). We further identify novel events such as complex 4-way admixture in San-Khomani individuals, and show that Eastern European populations possess 1 – 5% ancestry from a group resembling modern-day central Asians. We also identify evidence of recent natural selection favouring sub-Saharan ancestry at the HLA region, across North African individuals. We make available an R and C++ software library, which we term MOSAIC (which stands for MOSAIC Organises Segments of Ancestry In Chromosomes).