Search

The law does not concern itself with trifles. If a risk is deemed minimal, or an infraction negligible, invoking the authority of the law often seems unnecessary. However, there are increasingly fields of human activity where this principle leads to gaps in the protection necessary for a flourishing society. This paper reports findings and ideas from a research project in cumulative data disclosure, where an aggregation of in themselves harmless data points can expose the users of social media to significant personal risk.

Small, apparently innocuous pieces of personal data, generated during online interactions, can, over time, collectively pose significant yet often unanticipated risks to personal reputation, and to employers' operational security. Such revelations arise from personal data shared intentionally by individuals, data shared about them by others, from tagging software, and from everyday cross-authentication practices. When brought together, unintended insights into personal habits, work patterns, emotions and political leanings can be gained by others, potentially creating adverse consequences for individuals e.g. via reputational damage, employers e.g. via cybercrime, and even for national security. Our research examines potential adverse revelations resulting from individual digital traces via the development of a tool that provides portraits of digital traces and reflects them back to the user. The tool will enable individuals to understand the cumulative nature of their data, and to better comprehend the resultant vulnerabilities. Our research also aims to identify conflict and ambiguity in cybersecurity service design and implementation when motivations and actions of employees are balanced against organisational security philosophy. This will enable mitigation against risks, issues and consequences of cumulative revelations from both an organisational and individual perspective. We focus here on an interview study conducted with 26 participants during the UK COVID-19 lockdown, examining their digital ecosystems, practices and experiences of cumulative revelations, where combined elements of their personal data revealed more than intended. We discuss individual awareness and perception of online presence, events that shape individual online behaviour, unanticipated sources of risk, and data privacy literacy. We discuss the implications of cumulative revelations for individuals and employers and the intersection with COVID-related homeworking on online practices and

This paper briefly describes our research groups' efforts in tackling Task 1 (Early Detection of Signs of Self-Harm), and Task 2 (Measuring the Severity of the Signs of Depression) from the CLEF eRisk Track. Core to how we approached these problems was the use of BERT-based classifiers which were trained specifically for each task. Our results on both tasks indicate that this approach delivers high performance across a series of measures, particularly for Task 1, where our submissions obtained the best performance for precision, F1, latency-weighted F1 and ERDE at 5 and 50. This work suggests that BERT-based classifiers, when trained appropriately, can accurately infer which social media users are at risk of self-harming, with precision up to 91.3% for Task 1. Given these promising results, it will be interesting to further refine the training regime, classifier and early detection scoring mechanism, as well as apply the same approach to other related tasks (e.g., anorexia, depression, suicide).

Sentiment analysis is central to the process of mining opinions and attitudes from online texts. While much attention has been paid to the sentiment classification problem, much less work has tried to tackle the problem of predicting the intensity of the sentiment. The go to method is VADER --- an unsupervised lexicon based approach to scoring sentiment. However, such approaches are limited because of the vocabulary mismatch problem. In this paper, we present in detail and evaluate our AWESSOME framework (A Word Embedding Sentiment Scorer Of Many Emotions) for sentiment intensity scoring, that capitalizes on pre-existing lexicons, does not require training and provides fine grained and accurate sentiment intensity scores of words, phrases and text. In our experiments, we used seven Sentiment Collections to evaluate the proposed approach, against lexicon based approaches (e.g., VADER), and supervised methods such as deep learning based approaches (e.g., SentiBERT). The results show that despite not surpassing supervised approaches, the AWESSOME unsupervised approach significantly outperforms existing lexicon approaches and therefore provides a simple and effective approach for sentiment analysis. The AWESSOME framework can be flexibly adapted to cater for different seed lexicons and different neural word embeddings models in order to produce corpus specific lexicons -- without the need for extensive supervised learning or retraining.

This paper briefly describes our research groups’ efforts in tackling Task 1 (Early Detection of Signs of Self-Harm), and Task 2 (Measuring the Severity of the Signs of Depression) from the CLEF eRisk Track. Core to how we approached these problems was the use of BERT-based classifiers which were trained specifically for each task. Our results on both tasks indicate that this approach delivers high performance across a series of measures, particularly for Task 1, where our submissions obtained the best performance for precision, F1, latency-weighted F1 and ERDE at 5 and 50. This work suggests that BERT-based classifiers, when trained appropriately, can accurately infer which social media users are at risk of self-harming, with precision up to 91.3% for Task 1. Given these promising results, it will be interesting to further refine the training regime, classifier and early detection scoring mechanism, as well as apply the same approach to other related tasks (e.g., anorexia, depression, suicide).

We present, in this paper, our contribution in DEFT 2018 task 2 : "Global polarity", determining the overall polarity (Positive, Negative, Neutral or MixPosNeg) of tweets regarding public transport, in French language. Our system is based on a list of sentiment seed-words adapted for French public transport tweets. These seed-words are extracted from DEFT's training annotated dataset, and the sentiment relations between seed-words and other terms are captured by cosine measure of their word embeddings representations, using a French language word embeddings model of 683k words. Our semi-supervised system achieved an F1-measure equals to 0.64. RÉSUMÉ Mots-graines de Similarité de Sentiment Adaptés pour la Classification de Polarité des Tweets en Langue Française. Cet article présente notre contribution en DEFT 2018 tâche 2 : "Polarité globale", déterminant la polarité globale (Positif, Négatif, Neutre ou MixPosNeg) des tweets concernant les transports publics, en langue Française. Notre système est basé sur une liste de mots-graines de sentiment adaptés aux tweets de transport public français. Ces mots-graines sont extraits de corpus annoté de DEFT, et les relations entre les mots-graine et les autres termes sont capturées par la similarité en mesure de cosinus entre les vecteurs représentants les mots, en utilisant un modèle word2vec en langue Française de 683k mots. Notre système semi-supervisé a atteint un F1-measure égale à 0,64.

We present, in this paper, our contribution in SemEval2017 task 4 : " Sentiment Analysis in Twitter " , subtask A: " Message Polarity Classification " , for En-glish and Arabic languages. Our system is based on a list of sentiment seed words adapted for tweets. The sentiment relations between seed words and other terms are captured by cosine similarity between the word embedding representations (word2vec). These seed words are extracted from datasets of annotated tweets available online. Our tests, using these seed words, show significant improvement in results compared to the use of Turney and Littman's (2003) seed words, on polarity classification of tweet messages.

In this paper, we present the automatic annotation of bibliographical references' zone in papers and articles of XML/TEI format. Our work is applied through two phases: first, we use machine learning technology to classify bibliographical and non-bibliographical paragraphs in papers, by means of a model that was initially created to differentiate between the footnotes containing or not containing bibliographical references. The previous description is one of BILBO's features, which is an open source software for automatic annotation of bibliographic reference. Also, we suggest some methods to minimize the margin of error. Second, we propose an algorithm to find the largest list of bibliographical references in the article. The improvement applied on our model results an increase in the model's efficiency with an Accuracy equal to 85.89. And by testing our work, we are able to achieve 72.23% as an average for the percentage of success in detecting bibliographical references' zone.

In this paper, we present our contribution in SemEval2016 task7 1 : Determining Sentiment Intensity of English and Arabic Phrases, where we use web search engines for English and Arabic unsupervised sentiment intensity prediction. Our work is based, first, on a group of classic sentiment lexicons (e.g. Sen-timent140 Lexicon, SentiWordNet). Second, on web search engines' ability to find the co-occurrence of sentences with predefined negative and positive words. The use of web search engines (e.g. Google Search API) enhance the results on phrases built from opposite polarity terms.