Enron Email Corpus Research Articles

ParaGuide: Guided Diffusion Paraphrasers for Plug-and-Play Textual Style Transfer

Textual style transfer is the task of transforming stylistic properties of text while preserving meaning. Target "styles" can be defined in numerous ways, ranging from single attributes (e.g. formality) to authorship (e.g. Shakespeare). Previous unsupervised style-transfer approaches generally rely on significant amounts of labeled data for only a fixed set of styles or require large language models. In contrast, we introduce a novel diffusion-based framework for general-purpose style transfer that can be flexibly adapted to arbitrary target styles at inference time. Our parameter-efficient approach, ParaGuide, leverages paraphrase-conditioned diffusion models alongside gradient-based guidance from both off-the-shelf classifiers and strong existing style embedders to transform the style of text while preserving semantic information. We validate the method on the Enron Email Corpus, with both human and automatic evaluations, and find that it outperforms strong baselines on formality, sentiment, and even authorship style transfer.

Stylistic variation in email

Abstract This study explores how email is partly shaped by writers’ positions within a corporate structure. This stylistic variation is measurable at scale and can be described by messages’ rhetorical organizations and orientations. The modeling was carried out on a subset of the Enron email corpus, which was processed using the dictionary-based tagger DocuScope. The results identify four stylistic variants (Trained/Technical Support, Decision-Making, Everyday Workplace Interaction, and Engaged Planning), each realizing distinctive combinations of features reflective of their communicative functions. In Trained/Technical Support emails, for example, constellations of words and phrases associated with informational production and facilitation are marshaled in fulfilling routine guidance-seeking and guidance-giving tasks. While writers’ positions motivate stylistic tendencies (e.g., members of upper-level management compose a majority of their messages in the Decision-Making style), all writers avail themselves of a variety of styles, depending on audience and purpose, suggesting that learners might benefit from developing adaptable communicative repertoires.

Latent Set Models for Two-Mode Network Data

Two-mode networks are a natural representation for many kinds of relational data. These networks are bipartite graphs consisting of two distinct sets ("modes") of entities. For example, one can model multiple recipient email data as a two-mode network of (a) individuals and (b) the emails that they send or receive. In this work we present a statistical model for two-mode network data which posits that individuals belong to latent sets and that the members of a particular set tend to co-appear. We show how to infer these latent sets from observed data using a Markov chain Monte Carlo inference algorithm. We apply the model to the Enron email corpus, using it to discover interpretable latent structure as well as evaluating its predictive accuracy on a missing data task. Extensions to the model are discussed that incorporate additional side information such as the email's sender or text content, further improving the accuracy of the model.

Non‐parametric regression for networks

Network data are becoming increasingly available, and so there is a need to develop a suitable methodology for statistical analysis. Networks can be represented as graph Laplacian matrices, which are a type of manifold‐valued data. Our main objective is to estimate a regression curve from a sample of graph Laplacian matrices conditional on a set of Euclidean covariates, for example, in dynamic networks where the covariate is time. We develop an adapted Nadaraya–Watson estimator which has uniform weak consistency for estimation using Euclidean and power Euclidean metrics. We apply the methodology to the Enron email corpus to model smooth trends in monthly networks and highlight anomalous networks. Another motivating application is given in corpus linguistics, which explores trends in an author's writing style over time based on word co‐occurrence networks.

Employee profiling via aspect-based sentiment and network for insider threats detection

Historically, the harm caused by insiders has proven to be one of the greatest concerns for any organization. As such, it has received considerable attention from both the industrial and research communities. Existing works mainly focused on modeling the employees’ normal biometric behavior (e.g., human to device interaction pattern) to detect anomalous behavior which corresponds to the insider activity. However, it is unattainable to stop the insider at the final moment when the malicious act is being carried out. In this paper, we propose a novel framework which performs employee profiling based on aspect-based sentiments and social network information and examine its applicability for early detection of potential insider threats. On the contrary to the traditional sentiment analysis, aspect-based sentiment analysis provides more fine-grained information on the employee. Our framework employs a combination of deep learning techniques such as Gated Recurrent Unit (GRU) and skipgram to build temporal sentiment profiles for the employees. It then performs anomaly detection on the profiles and ranks the employees based on their respective anomaly score. Due to the absence of relevant benchmark dataset, we augmented the publicly available real-world Enron email corpus with an insider threat scenario to evaluate our framework. The evaluation results demonstrate that the augmentation is indeed reflected in the augmented employee’s anomaly ranking (i.e., from normal to abnormal) and her close associates are indeed placed closely to her when the profiles are visualized in the 2D space. The profiles obtained from our framework can also be used to complement any existing expert and intelligent systems with additional capabilities in handling textual information such as, integration with profiles obtained from biometric behavior to form a more comprehensive threat detection system.

An EWMA control chart for categorical processes with applications to social network monitoring

Statistical process monitoring (SPM) has traditionally been applied to manufacturing processes; however, more recent developments highlight its application to social networks. We propose a network monitoring strategy using the exponentially weighted moving average (EWMA) control chart, with the goal of detecting shifts in the hierarchical tendency of directed graphs over time. Such a strategy might prove useful to an organization’s stakeholders when interest lies in monitoring for shifts in the general health of the organization. Although development of the proposed control chart was motivated by a network monitoring problem, our method is generally applicable to multinomial categorical processes. We study the detection performance of our proposed control chart, relative to that of the multinomial cumulative sum (CUSUM) alternative. Results suggest that if the out-of-control shift in the multinomial probabilities cannot be specified a priori, the proposed EWMA control chart should be considered as an alternative to the CUSUM strategy. Application of the proposed method is demonstrated on a real data set, namely, the open-source Enron e-mail corpus.

Sentiment analysis in organizational work: Towards an ontology of people analytics

AbstractThe present paper proposes a conceptual ontology to evaluate human factors by modelling their key performance indicators and defining these indicators' explanatory factors, manifestations, and diverse corresponding digital footprints. Our methodology incorporates 6 main human resource constructs: performance, engagement, leadership, workplace dynamics, organizational developmental support, and learning and knowledge creation. Using sentiment analysis, we introduce a potential way to evaluate several components of the proposed human factors ontology. We use the Enron email corpus as a test case, to demonstrate how digital footprints can predict such phenomena. In so doing, we hope to encourage further research applying data mining techniques to allow real‐time, less costly, and more reliable assessments of human factor patterns and trends.

Managerial hubris detection: the case of Enron

Hubris is a known risk for leadership failure. We show that hubristic tendencies can be detected semantically ex-ante in textual reports, and offer a novel methodology aimed at detecting real-time hubristic propensities. The methodology employs text mining based on natural language processing (NLP) on Enron email corpus. NLP can capture information about employees and predict change patterns. Employing NLP real-time mechanism, Enron executives’ hubristic tendencies were detected. Findings indicate that hubristic expressions amongst senior executives are significantly more frequent than amongst their non-senior counterparts, and that the frequency of hubristic expressions increases the closer one gets to Enron’s collapse. Whilst both Enron’s CEO’s were hubristic, we found Skilling to be typified with severer hubris. Our study is the first to employ NLP real-time analytical process to detect the hubris disposition. Predicated on Enron’s case study, we demonstrate the methodology’s strengths, notably immediate recognition of accumulated symptoms and prevalence.

Improving corpus reproducibility through modular text transformations and connected data set

ABSTRACTThe Enron Email Corpus is one of the most utilized collections of documents in Natural Language Processing, Machine Learning, and Network Analysis. Different groups of researchers have transformed the corpus, changing the content and format to meet their needs. The many distinct versions can all claim to be the Enron Email Corpus, though they are as distinct from the original publicly available collection as they are from each other. Researchers then have to determine the usefulness of a particular version in comparison to the many others available, as well as ascertain what has been done to the collection and how it would affect their specific research goal. This is especially important for reproducing a particular research method onto a different corpus, as transposing a method necessitates a deep understanding of the original data in the experiment. This project models the various transformations performed on different versions of the collection to form a network of connected datasets, highlighting the most important nodes as well as the most common transformations. Traversing different paths between nodes offers the community a way to model and reproduce the necessary data work performed on one collection that can be transposed onto other collections.

Using word n-grams to identify authors and idiolects

AbstractForensic authorship attribution is concerned with identifying the writers of anonymous criminal documents. Over the last twenty years, computer scientists have developed a wide range of statistical procedures using a number of different linguistic features to measure similarity between texts. However, much of this work is not of practical use to forensic linguists who need to explain in reports or in court why a particular method of identifying potential authors works. This paper sets out to address this problem using a corpus linguistic approach and the 176-author 2.5 million-word Enron Email Corpus. Drawing on literature positing the idiolectal nature of collocations, phrases and word sequences, this paper tests the accuracy of word n-grams in identifying the authors of anonymised email samples. Moving beyond the statistical analysis, the usage-based concept of entrenchment is offered as a means by which to account for the recurring and distinctive production of idiolectal word n-grams.

CommViz: Visualization of semantic patterns in large social communication networks

We introduce CommViz, an information visualization tool that enables complex communication networks to be explored, exposing trends and patterns in the data at a glance. We adapt a visualization approach known as hive plots to reflect the semantic structure of the networks, a generalization we call semantic hive plots. The method efficiently organizes and provides insight into complex, high-dimensional communication data such as email and messages on social media. We present the architecture of the CommViz tool and its application to the Enron email corpus as a case study, demonstrating how the structure of the visualization enables investigation of patterns and relationships in a large set of messages. We also provide a user study performed with Amazon Mechanical Turk that shows the value of the tool for certain complex data interrogations and further show how the incorporation of semantic structure on semantic coordinates can also be applied to parallel coordinates visualization. The integration of the social network characteristics with semantic attributes of the underlying data in a single visualization is, to our knowledge, a novel contribution of the work. The tool can be accessed at http://commviz.eng.unimelb.edu.au . Code is available at https://bitbucket.org/readbiomed/commviz . The Enron email corpus is available from http://bailando.sims.berkeley.edu/enron_email.html .

Sharing Social Network Data: Differentially Private Estimation of Exponential Family Random-Graph Models

SummaryMotivated by a real life problem of sharing social network data that contain sensitive personal information, we propose a novel approach to release and analyse synthetic graphs to protect privacy of individual relationships captured by the social network while maintaining the validity of statistical results. A case-study using a version of the Enron e-mail corpus data set demonstrates the application and usefulness of the proposed techniques in solving the challenging problem of maintaining privacy and supporting open access to network data to ensure reproducibility of existing studies and discovering new scientific insights that can be obtained by analysing such data. We use a simple yet effective randomized response mechanism to generate synthetic networks under ε-edge differential privacy and then use likelihood-based inference for missing data and Markov chain Monte Carlo techniques to fit exponential family random-graph models to the generated synthetic networks.

Network Analysis with the Enron Email Corpus

We use the Enron email corpus to study relationships in a network by applying six different measures of centrality. Our results came out of an in-semester undergraduate research seminar. The Enron corpus is well suited to statistical analyses at all levels of undergraduate education. Through this article's focus on centrality, students can explore the dependence of statistical models on initial assumptions and the interplay between centrality measures and hierarchical ranking, and they can use completed studies as springboards for future research. The Enron corpus also presents opportunities for research into many other areas of analysis, including social networks, clustering, and natural language processing.

Online Convex Optimization in Dynamic Environments

High-velocity streams of high-dimensional data pose significant “big data” analysis challenges across a range of applications and settings. Online learning and online convex programming play a significant role in the rapid recovery of important or anomalous information from these large datastreams. While recent advances in online learning have led to novel and rapidly converging algorithms, these methods are unable to adapt to nonstationary environments arising in real-world problems. This paper describes a dynamic mirror descent framework which addresses this challenge, yielding low theoretical regret bounds and accurate, adaptive, and computationally efficient algorithms which are applicable to broad classes of problems. The methods are capable of learning and adapting to an underlying and possibly time-varying dynamical model. Empirical results in the context of dynamic texture analysis, solar flare detection, sequential compressed sensing of a dynamic scene, traffic surveillance, tracking self-exciting point processes and network behavior in the Enron email corpus support the core theoretical findings.

Forensic triage of email network narratives through visualisation

Purpose – The purpose of this paper is to propose a novel approach that automates the visualisation of both quantitative data (the network) and qualitative data (the content) within emails to aid the triage of evidence during a forensics investigation. Email remains a key source of evidence during a digital investigation, and a forensics examiner may be required to triage and analyse large email data sets for evidence. Current practice utilises tools and techniques that require a manual trawl through such data, which is a time-consuming process. Design/methodology/approach – This paper applies the methodology to the Enron email corpus, and in particular one key suspect, to demonstrate the applicability of the approach. Resulting visualisations of network narratives are discussed to show how network narratives may be used to triage large evidence data sets. Findings – Using the network narrative approach enables a forensics examiner to quickly identify relevant evidence within large email data sets. Within the case study presented in this paper, the results identify key witnesses, other actors of interest to the investigation and potential sources of further evidence. Practical implications – The implications are for digital forensics examiners or for security investigations that involve email data. The approach posited in this paper demonstrates the triage and visualisation of email network narratives to aid an investigation and identify potential sources of electronic evidence. Originality/value – There are a number of network visualisation applications in use. However, none of these enable the combined visualisation of quantitative and qualitative data to provide a view of what the actors are discussing and how this shapes the network in email data sets.

Locality Statistics for Anomaly Detection in Time Series of Graphs

The ability to detect change-points in a dynamic network or a time series of graphs is an increasingly important task in many applications of the emerging discipline of graph signal processing. This paper formulates change-point detection as a hypothesis testing problem in terms of a generative latent position model, focusing on the special case of the Stochastic Block Model time series. We analyze two classes of scan statistics, based on distinct underlying locality statistics presented in the literature. Our main contribution is the derivation of the limiting properties and power characteristics of the competing scan statistics. Performance is compared theoretically, on synthetic data, and empirically, on the Enron email corpus.

Quantifying and Comparing Centrality Measures for Network Individuals as Applied to the Enron Corpus

The ever increasing body of social networks creates an opportunity for extensive network analysis and investigations of communications, cliques, and network contributions. In this study, we focus our attention on the Enron email corpus and the corresponding network of employees, attempting to gather information from the email communications. Methods of data reduction on the email corpus were used to create a weighted adjacency matrix in which each i, j-entry corresponds to a weighted count of correspondences from employee i to employee j. While there are many ways to measure importance within a corporate network, of which job title constitutes one such measure, our study focuses on five primary measures: eigenvector centrality, row-sums of a topological overlap matrix, closeness, betweenness, and Opsahl metric. These network analysis metrics were applied to the weighted adjacency matrix to calculate the centrality measures for each individual employee, which were subsequently compiled into ordinally ranked lists of employees for each centrality measure based on decreasing importance. Additionally, the centrality data was visualized using the DataDriven Documents (D3) javascript library, allowing for network visualization in terms of department job title and number of emails sent. In applying the centrality measures to network data, we explore the differences inherent in each measure and work to compare them as well as the corresponding employee importance rankings for each. The metrics in our analysis determined individual importance of employees by applying significant weight to various aspects of the employees’ network roles. By identifying employees that are connected to a large number of individuals and simultaneously have extensive correspondences with those individuals, the Opsahl score combines the other measures, proving to be the most useful metric in exploring Enron’s inner-corporate structure.

Factors Affecting Effective Organizational Communication?

Organizational communication is key to the act of organizing and accomplishment of organizational goals. Previous research shows that simplicity is associated with more efficient communication. In this study, we investigate factors that may lead individuals in organizations to deviate from the use of simple communication. In a study of the Enron email corpus, we find that the hierarchical direction (upward or downward) of the communication, hierarchical distance between communicators, and size of the audience are associated with changes in the level of simplicity (readability) of email communication in ways which are contrary to what may be deemed effective.

D-HOCS: an algorithm for discovering the hierarchical overlapping community structure of a social network

Social networks often demonstrate a hierarchical organization, with communities embedded within other communities; moreover, nodes can be shared between different communities, i.e. communities in social networks may be overlapping. In this paper, we define a hierarchical overlapping community structure to present overlapping communities of a social network at different levels of granularity. Discovering the hierarchical overlapping community structure of a social network can provide us a deeper understanding of the complex nature of social networks. We propose an algorithm, called D-HOCS, to derive the hierarchical overlapping community structure of social networks. Firstly, D-HOCS generates a probability transition matrix by applying random walk to a social network, and then trains a Gaussian Mixture Model using the matrix. Further D-HOCS derives overlapping communities by analyzing mean vectors of the Gaussian mixture model. Varying the number of components, D-HOCS repeatedly trains the Gaussian mixture model, detecting the overlapping communities at different levels of granularity. Organizing the overlapping communities into a hierarchy, D-HOCS can finally obtain the hierarchical overlapping community structure of the social network. The experiments conducted on synthetic and real dataset demonstrate the feasibility and applicability of the proposed algorithm. We further employ D-HOCS to explore Enron e-mail corpus, and obtain several interesting insights. For example, we find out a coordinator who coordinated many sections of the Enron Corporation to complete an important task during first half of 2001. We also identify a community that corresponds to a real organization in Enron Corporation.

Stylistic variation within genre conventions in the Enron email corpus: developing a textsensitive methodology for authorship research

Over recent years there has been much theoretical discussion regarding idiolect and its usefulness in forensic authorship analysis. This article, drawing on email data from the former American energy company Enron, offers an empirical investigation into identifying individuals’ idiolects through analysing author distinctive variation within two conventions of the email genre – greetings and farewells. The first part of a two-stage analysis identifies a number of forms which distinguish between authors in a four-author corpus. Using likelihood ratios, the second stage of analysis finds that some of the greeting and farewell forms identified, and combinations of forms, remain distinctive and individuating of their author when tested against the 126-author ‘Enron Sent Email Author Reference Corpus’, and highlights the diagnostic power of less frequent variants. The results from this article offer both theoretical and methodological contributions as well as a baseline of population results for forensic authorship casework involving emails.