We investigate the problem of the robustness of a trained neural network to the perturbation of its input values. More specifically, we examine the interplay between the accuracy of the network, as measured by the mean squared error, and robustness. Accordingly, we present a robustness measure, which, with high probability, suggests an upper bound on the mean squared error of the network, with respect to an input data set, for a given perturbation of the input values of the network. The measure we propose is both simple and efficient to compute, treating the neural network as a black box. We provide experimental results on several real-world data sets showing the efficacy of the proposed method. We also introduce the concept of robustness curves, which allows us to further analyse robustness within and between data sets.

In this paper, we investigate the dynamics of the social media response on Reddit to the COVID-19 pandemic during its first year (February 2020-2021). The emergence of region-specific subreddits allows us to compare the reactions of individuals posting their opinions on social media about the global pandemic from two perspectives - the UK and the US.In particular, we look at the volume of posts and comments on these two subreddits, and at the sentiment expressed in these posts and comments over time as a measure of the public level of engagement and response. Whilst an analysis of volume allows us to quantify how interested people are about the pandemic as it unfolds, sentiment analysis goes beyond this and informs us about how people respond towards the pandemic based on the textual content in the posts and comments. The research looks to develop a framework for analyzing the social response on Reddit to a large-scale event in terms of the level of engagement measured through post and comment volumes, and opinion measured through an analysis of sentiment applied to the post content. In order to compare the subreddits, we show the trend in the time series through the application of moving average methods. We also show how to identify the lag between time series and align them using cross-correlation. Moreover, once aligned, we apply moving correlations to the time series to measure their degree of correspondence to see if there is a similar response to the pandemic across the two groups (UK and US). The results indicate that both subreddits were posting in high volumes at specific points during the pandemic, and that, despite the generally negative sentiment in the posts and comments, a gradual decrease in negativity leading up to the start of 2021 is observed as measures are put in place by governments and organizations to contain the virus and provide necessary support the affected populations.

We investigate the potential benefit of incorporating dictionary information into a neural network architecture for natural language processing. In particular, we make use of this architecture to extract several concepts related to COVID-19 from an on-line medical forum. We use a sample from the forum to manually curate one dictionary for each concept. In addition, we use MetaMap, which is a tool for extracting biomedical concepts, to identify a small number of semantic concepts. For a supervised concept extraction task on the forum data, our best model achieved a macro F_1 score of 90%. A major difficulty in medical concept extraction is obtaining labelled data from which to build supervised models. We investigate the utility of our models to transfer to data derived from a different source in two ways. First for producing labels via weak learning and second to perform concept extraction. The dataset we use in this case comprises COVID-19 related tweets and we achieve an F_1 score 81% for symptom concept extraction trained on weakly labelled data. The utility of our dictionaries is compared with a COVID-19 symptom dictionary that was constructed directly from Twitter. Further experiments that incorporate BERT and a COVID-19 version of BERTweet demonstrate that the dictionaries provide a commensurate result. Our results show that incorporating small domain dictionaries to deep learning models can improve concept extraction tasks. Moreover, models built using dictionaries generalize well and are transferable to different datasets on a similar task.

Random minimaxing studies the consequences of using a random number for scoring the leaf nodes of a full width game tree and then computing the best move using the standard minimax procedure. Experiments in Chess showed that the strength of play increases as the depth of the lookahead is increased. Previous research by the authors provided a partial explanation of why random minimaxing can strengthen play by showing that, when one move dominates another move, then the dominating move is more likely to be chosen by minimax. This paper examines a special case of determining the move probability when domination does not occur. Specifically, we show that, under a uniform branching game tree model, whether the probability that one move is chosen rather than another depends not only on the branching factors of the moves involved, but also on whether the number of ply searched is odd or even. This is a new type of game tree pathology, where the minimax procedure will change its mind as to which move is best, independently of the true value of the game, and oscillate between moves as the depth of lookahead alternates between odd and even.

Branching processes are discrete-time stochastic processes which have been largely employed to model and simulate information diffusion processes over large online social networks such as Twitter and Reddit. Here we show that a variant of the branching process model enables the prediction of the popularity of user-generated content and thus can serve as a method for ranking search results or suggestions displayed to users. The proposed branching-process variant is able to evaluate the importance of an agent in a social network and, thus we propose a novel centrality index, called the Stochastic Potential Gain (SPG). The SPG is the first centrality index which combines the knowledge of the network topology with a dynamic process taking place on it which we call a graph-driven branching process. SPG generalises a range of popular network centrality metrics such as Katz' and Subgraph. We formulate a Monte Carlo algorithm (called MCPG) to compute the SPG and prove that it is convergent and correct. Experiments on two real datasets drawn from Facebook and GitHub demonstrate that MCPG traverses only a small fraction of nodes to produce its result, thus making the Stochastic Potential Gain an appealing option to compute node centrality measure for Online social networks.

A novel yet simple extension of the symmetric logistic distribution is proposed by introducing a skewness parameter. It is shown how the three parameters of the ensuing skew logistic distribution may be estimated using maximum likelihood. The skew logistic distribution is then extended to the skew bi-logistic distribution to allow the modelling of multiple waves in epidemic time series data. The proposed skew-logistic model is validated on COVID-19 data from the UK, and is evaluated for goodness-of-fit against the logistic and normal distributions using the recently formulated empirical survival Jensen–Shannon divergence (ESJS) and the Kolmogorov–Smirnov two-sample test statistic (KS2). We employ 95% bootstrap confidence intervals to assess the improvement in goodness-of-fit of the skew logistic distribution over the other distributions. The obtained confidence intervals for the ESJS are narrower than those for the KS2 on using this dataset, implying that the ESJS is more powerful than the KS2.

BACKGROUND:The COVID-19 pandemic has created a pressing need for integrating information from disparate sources in order to assist decision makers. Social media is important in this respect; however, to make sense of the textual information it provides and be able to automate the processing of large amounts of data, natural language processing methods are needed. Social media posts are often noisy, yet they may provide valuable insights regarding the severity and prevalence of the disease in the population. Here, we adopt a triage and diagnosis approach to analyzing social media posts using machine learning techniques for the purpose of disease detection and surveillance. We thus obtain useful prevalence and incidence statistics to identify disease symptoms and their severities, motivated by public health concerns.OBJECTIVE:This study aims to develop an end-to-end natural language processing pipeline for triage and diagnosis of COVID-19 from patient-authored social media posts in order to provide researchers and public health practitioners with additional information on the symptoms, severity, and prevalence of the disease rather than to provide an actionable decision at the individual level.METHODS:The text processing pipeline first extracted COVID-19 symptoms and related concepts, such as severity, duration, negations, and body parts, from patients' posts using conditional random fields. An unsupervised rule-based algorithm was then applied to establish relations between concepts in the next step of the pipeline. The extracted concepts and relations were subsequently used to construct 2 different vector representations of each post. These vectors were separately applied to build support vector machine learning models to triage patients into 3 categories and diagnose them for COVID-19.RESULTS:We reported macro- and microaveraged F1 scores in the range of 71%-96% and 61%-87%, respectively, for the triage and diagnosis of COVID-19 when the models were trained on human-labeled data. Our experimental results indicated that similar performance can be achieved when the models are trained using predicted labels from concept extraction and rule-based classifiers, thus yielding end-to-end machine learning. In addition, we highlighted important features uncovered by our diagnostic machine learning models and compared them with the most frequent symptoms revealed in another COVID-19 data set. In particular, we found that the most important features are not always the most frequent ones.CONCLUSIONS:Our preliminary results show that it is possible to automatically triage and diagnose patients for COVID-19 from social media natural language narratives, using a machine learning pipeline in order to provide information on the severity and prevalence of the disease for use within health surveillance systems.

A novel yet simple extension of the symmetric logistic distribution is proposed by introducing a skewness parameter. It is shown how the three parameters of the ensuing skew logistic distribution may be estimated using maximum likelihood. The skew logistic distribution is then extended to the skew bi-logistic distribution to allow the modelling of multiple waves in epidemic time series data. The proposed model is validated on COVID-19 data from the UK.

The coefficient of determination, known as $R^2$, is commonly used as a goodness-of-fit criterion for fitting linear models. $R^2$ is somewhat controversial when fitting nonlinear models, although it may be generalised on a case-by-case basis to deal with specific models such as the logistic model. Assume we are fitting a parametric distribution to a data set using, say, the maximum likelihood estimation method. A general approach to measure the goodness-of-fit of the fitted parameters, which is advocated herein, is to use a nonparametric measure for comparison between the empirical distribution, comprising the raw data, and the fitted model. In particular, for this purpose we put forward the Survival Jensen-Shannon divergence ($SJS$) and its empirical counterpart (${\cal E}SJS$) as a metric which is bounded, and is a natural generalisation of the Jensen-Shannon divergence. We demonstrate, via a straightforward procedure making use of the ${\cal E}SJS$, that it can be used as part of maximum likelihood estimation or curve fitting as a measure of goodness-of-fit, including the construction of a confidence interval for the fitted parametric distribution. Furthermore, we show the validity of the proposed method with simulated data, and three empirical data sets.

A bootstrap-based hypothesis test of the goodness-of-fit for the marginal distribution of a time series is presented. Two metrics, the empirical survival Jensen–Shannon divergence (ESJS) and the Kolmogorov–Smirnov two-sample test statistic (KS2), are compared on four data sets—three stablecoin time series and a Bitcoin time series. We demonstrate that, after applying first-order differencing, all the data sets fit heavy-tailed α-stable distributions with 1<α<2 at the 95% confidence level. Moreover, ESJS is more powerful than KS2 on these data sets, since the widths of the derived confidence intervals for KS2 are, proportionately, much larger than those of ESJS.

BACKGROUND The COVID-19 pandemic has created a pressing need for integrating information from disparate sources, in order to assist decision makers. Social media is important in this respect, however, to make sense of the textual information it provides and be able to automate the processing of large amounts of data, natural language processing methods are needed. Social media posts are often noisy, yet they may provide valuable insights regarding the severity and prevalence of the disease in the population. In particular, machine learning techniques for triage and diagnosis could allow for a better understanding of what social media may offer in this respect. OBJECTIVE This study aims to develop an end-to-end natural language processing pipeline for triage and diagnosis of COVID-19 from patient-authored social media posts, in order to provide researchers and other interested parties with additional information on the symptoms, severity and prevalence of the disease. METHODS The text processing pipeline first extracts COVID-19 symptoms and related concepts such as severity, duration, negations, and body parts from patients’ posts using conditional random fields. An unsupervised rule-based algorithm is then applied to establish relations between concepts in the next step of the pipeline. The extracted concepts and relations are subsequently used to construct two different vector representations of each post. These vectors are applied separately to build support vector machine learning models to triage patients into three categories and diagnose them for COVID-19. RESULTS We report that Macro- and Micro-averaged F_{1\ }scores in the range of 71-96% and 61-87%, respectively, for the triage and diagnosis of COVID-19, when the models are trained on human labelled data. Our experimental results indicate that similar performance can be achieved when the models are trained using predicted labels from concept extraction and rule-based classifiers, thus yielding end-to-end machine learning. Also, we highlight important features uncovered by our diagnostic machine learning models and compare them with the most frequent symptoms revealed in another COVID-19 dataset. In particular, we found that the most important features are not always the most frequent ones. CONCLUSIONS Our preliminary results show that it is possible to automatically triage and diagnose patients for COVID-19 from natural language narratives using a machine learning pipeline, in order to provide additional information on the severity and prevalence of the disease through the eyes of social media.

Objective: This study aims to develop an end-to-end natural language processing pipeline for triage and diagnosis of COVID-19 from patient-authored social media posts, in order to provide researchers and public health practitioners with additional information on the symptoms, severity and prevalence of the disease rather than to provide an actionable decision at the individual level. Materials and Methods: The text processing pipeline first extracts COVID-19 symptoms and related concepts such as severity, duration, negations, and body parts from patients' posts using conditional random fields. An unsupervised rule-based algorithm is then applied to establish relations between concepts in the next step of the pipeline. The extracted concepts and relations are subsequently used to construct two different vector representations of each post. These vectors are applied separately to build support vector machine learning models to triage patients into three categories and diagnose them for COVID-19. Results: We report that macro- and micro-averaged F1 scores in the range of 71-96% and 61-87%, respectively, for the triage and diagnosis of COVID-19, when the models are trained on human labelled data. Our experimental results indicate that similar performance can be achieved when the models are trained using predicted labels from concept extraction and rule-based classifiers, thus yielding end-to-end machine learning. Also, we highlight important features uncovered by our diagnostic machine learning models and compare them with the most frequent symptoms revealed in another COVID-19 dataset. In particular, we found that the most important features are not always the most frequent ones.

Human dynamics and sociophysics build on statistical models that can shed light on and add to our understanding of social phenomena. We propose a generative model based on a stochastic differential equation that enables us to model the opinion polls leading up to the 2017 and 2019 UK general elections and to make predictions relating to the actual results of the elections. After a brief analysis of the time series of the poll results, we provide empirical evidence that the gamma distribution, which is often used in financial modelling, fits the marginal distribution of this time series. We demonstrate that the proposed poll-based forecasting model may improve upon predictions based solely on polls. The method uses the Euler–Maruyama method to simulate the time series, measuring the prediction error with the mean absolute error and the root mean square error, and as such could be used as part of a toolkit for forecasting elections.

We propose a two-dimensional bibliometric index that strikes a balance between quantity (as measured by the number of publications of a researcher) and quality (as measured by the number of citations to those publications). While the square of h-index is determined by the maximum area square that fits under the citation curve of an author when plotting the number of citations in decreasing order, the rec-index is determined by the maximum area rectangle that fits under the curve. In this context we may distinguish between authors with a few very highly-cited publications, who may have carried out some influential research, and prolific authors, who may have many publications but fewer citations per publication. The influence of a researcher may be measured via a restricted version of the rec-index, the recI-index, which is the maximum area vertical rectangle that fits under the citation curve. Similarly, the prolificity of a researcher may be measured via the recP-index, which is the maximum area horizontal rectangle that fits under the citation curve. This leads to the proposal of the two-dimensional bibliometric index (recI,recP), which captures both aspects of a researcher's output. We present a comprehensive empirical analysis of this two-dimensional index on two datasets: a large set of Google Scholar profiles (representing "typical" researchers) and a small set of Nobel prize winners. Our results demonstrate the potential of this two-dimensional index, since for both data sets there is a statistically significant number of researchers for whom recI is greater than recP In particular, for nearly 25% of the Google Scholar researchers and for nearly 60% of the Nobel prize winners, recI is greater than rec(P).

Our goal is to summarise and aggregate information from social media regarding the symptoms of a disease, the drugs used and the treatment effects both positive and negative. To achieve this we first apply a supervised machine learning method to automatically extract medical concepts from natural language text. In an environment such as social media, where new data is continuously streamed, we need a methodology that will allow us to continuously train with the new data. To attain such incremental re-training, a semi-supervised methodology is developed, which is capable of learning new concepts from a small set of labelled data together with the much larger set of unlabelled data. The semi-supervised methodology deploys a conditional random field (CRF) as the base-line training algorithm for extracting medical concepts. The methodology iteratively augments to the training set sentences having high confidence, and adds terms to existing dictionaries to be used as features with the base-line model for further classification. Our empirical results show that the base-line CRF performs strongly across a range of different dictionary and training sizes; when the base-line is built with the full training data the F1 score reaches the range 84%–90%. Moreover, we show that the semi-supervised method produces a mild but significant improvement over the base-line. We also discuss the significance of the potential improvement of the semi-supervised methodology and found that it is significantly more accurate in most cases than the underlying base-line model.

Human dynamics and sociophysics suggest statistical models that may explain and provide us with better insight into social phenomena. Here we tackle the problem of determining the distribution of the population density of a social space over time by modelling the dynamics of agents entering and exiting the space as a birth-death process. We show that, for a simple agent-based model in which the probabilities of entering and exiting the space depends on the number of agents currently present in the space, the population density of the space follows a gamma distribution. We also provide empirical evidence supporting the validity of the model by applying it to a data set of occupancy traces of a common space in an office building.

We propose a new word embedding model, called SPhrase, that incorporates supervised phrase information. Our method modifies traditional word embeddings by ensuring that all target words in a phrase have exactly the same context. We demonstrate that including this information within a context window produces superior embeddings for both intrinsic evaluation tasks and downstream extrinsic tasks.

Centrality metrics are a popular tool in Network Science to identify important nodes within a graph.We introduce the Potential Gain as a centrality measure that unifies many walk-based centrality metrics in graphs and captures the notion of node navigability, interpreted as the property of being reachable from anywhere else (in the graph) through short walks.Two instances of the Potential Gain (called the Geometric and the Exponential Potential Gain) are presented and we describe scalable algorithms for computing them on large graphs.We also give a proof of the relationship between the new measures and established centralities.The geometric potential gain of a node can thus be characterized as the product of its Degree centrality by its Katz centrality scores.At the same time, the exponential potential gain of a node is proved to be the product of Degree centrality by its Communicability index.These formal results connect potential gain to both the "popularity" and "similarity" properties that are captured by the above centralities.

PurposeThe purpose of this paper is to present a language-agnostic approach to facilitate the discovery of “parallel passages” stored in historic and cultural heritage digital archives.Design/methodology/approachThe authors explore a novel, and relatively simple approach, using a character-based statistical language model combined with a tailored version of the Basic Local Alignment Tool to extract exact and approximate string patterns shared between groups of documents.FindingsThe approach is applicable to a wide range of languages, and compensates for variability in the text of the documents as a result of differences in dialect, authorship, language change over time and errors due to inaccurate transcriptions and optical character recognition errors as a result of the digitisation process.Research limitations/implicationsA number of case studies demonstrate that the approach is practical and generalisable to a wide range of archives with documents in different languages, domains and of varying quality.Practical implicationsThe approach described can be applied to any digital archive of modern and contemporary texts. This makes the approach applicable to digital archives recording historic texts, but also those composed of more recent news articles, for example.Social implicationsThe analysis of “parallel passages” enables researchers to quantify the presence and extent of text-reuse in a collection of documents, which can provide useful data on author style, text genres and cultural contexts.Originality/valueThe approach is novel and addresses a need by humanities researchers for tools that can identify similar documents and local similarities represented by shared text sequences in a potentially vast large archive of documents. As far as the authors are aware, there are no tools currently exist that provide the same level of tolerance to the language of the documents.

Axiomatic characterisation of a bibliometric index provides insight into the properties that the index satisfies and facilitates the comparison of different indices. A geometric generalisation of the h-index, called the $$\chi$$-index, has recently been proposed to address some of the problems with the h-index, in particular, the fact that it is not scale invariant, i.e., multiplying the number of citations of each publication by a positive constant may change the relative ranking of two researchers. While the square of the h-index is the area of the largest square under the citation curve of a researcher, the square of the $$\chi$$-index, which we call the rec-index (or rectangle-index), is the area of the largest rectangle under the citation curve. Our main contribution here is to provide a characterisation of the rec-index via three properties: monotonicity, uniform citation and uniform equivalence. Monotonicity is a natural property that we would expect any bibliometric index to satisfy, while the other two properties constrain the value of the rec-index to be the area of the largest rectangle under the citation curve. The rec-index also allows us to distinguish between influential researchers who have relatively few, but highly-cited, publications and prolific researchers who have many, but less-cited, publications.