Search

The usage of social media has exponentially grown in recent years leaving the users with no limitations on misusing the platforms through abusive contents as deemed fit to them. This exacerbates abusive words exposure to innocent users, especially in social media forums, including children. In an attempt to alleviate the problem of abusive words proliferation on social media, researchers have proposed different methods to help deal with variants of the abusive words; however, obfuscated abusive words detection still poses challenges. A method that utilizes a combination of rule based approach and character percentage matching techniques is proposed to improve the detection rate for obfuscated abusive words. The evaluation results achieved F1 score percentage ratio of 0.97 and accuracy percentage ratio of 0.96 which were above the significance ratio of 0.5. Hence, the proposed approach is highly effective for obfuscated abusive words detection and prevention. Keywords: Rule based approach, Character percentage matching techniques, Obfuscated abuse, Abuse detection, Abusive words, Social media

User-generated web content is rife with abusive language that can harm others and discourage participation. Thus, a primary research aim is to develop abuse detection systems that can be used to alert and support human moderators of online communities. Such systems are notoriously hard to develop and evaluate. Even when they appear to achieve satisfactory performance on current evaluation metrics, they may fail in practice on new data. This is partly because datasets commonly used in this field suffer from selection bias, and consequently, existing supervised models overrely on cue words such as group identifiers (e.g., gay and black) which are not inherently abusive. Although there are attempts to mitigate this bias, current evaluation metrics do not adequately quantify their progress. In this work, we introduce Adversarial Attacks against Abuse (AAA), a new evaluation strategy and associated metric that better captures a model’s performance on certain classes of hard-to-classify microposts, and for example penalises systems which are biased on low-level lexical features. It does so by adversarially modifying the model developer’s training and test data to generate plausible test samples dynamically. We make AAA available as an easy-to-use tool, and show its effectiveness in error analysis by comparing the AAA performance of several state-of-the-art models on multiple datasets. This work will inform the development of detection systems and contribute to the fight against abusive language online.

International audience; With the spread of online social networks, it is more and more difficult to monitor all the user-generated content. Automating the moderation process of the inappropriate exchange content on Internet has thus become a priority task. Methods have been proposed for this purpose, but it can be challenging to find a suitable dataset to train and develop them. This issue is especially true for approaches based on information derived from the structure and the dynamic of the conversation. In this work, we propose an original framework, based on the Wikipedia Comment corpus, with comment-level abuse annotations of different types. The major contribution concerns the reconstruction of conversations, by comparison to existing corpora, which focus only on isolated messages (i.e. taken out of their conversational context). This large corpus of more than 380k annotated messages opens perspectives for online abuse detection and especially for context-based approaches. We also propose, in addition to this corpus, a complete benchmarking platform to stimulate and fairly compare scientific works around the problem of content abuse detection, trying to avoid the recurring problem of result replication. Finally, we apply two classification methods to our dataset to demonstrate its potential.

Extended abstract présenté lors de la conférence MARAMI 2020.; National audience; In recent years, online social media have allowed people to meet and discuss world-wide. These popular platforms are confronted with increasing abusive content. In order to automate the detection of abusive content in such social media, researchers have proposed various methods based on Natural Language Processing (NLP), and have leveraged behavioral information about users and the structure of conversations. In our previous work, we proposed to combine NLP and conversational graph-based features to detect abusive messages in chat logs extracted from an online game. These conversational graphs model interactions between users (i.e. who is arguing with whom?), while completely ignoring the language content of the messages. We characterized the structure of these graphs by computing a large set of manually selected topological measures, and used them as features to train a classifier into detecting abusive messages. Graph embedding methods allow representing graphs as low-dimensional vectors while preserving at least a part of their topological properties. In addition to the plain structure, certain methods are able to capture additional information such as node labels or the weight and direction of edges. These representations are automatically learned, so they have the advantage of not requiring to perform any feature selection or feature engineering. One can distinguish four main categories of graph embedding methods, depending on the nature of the considered objects: node, edge, subgraph and whole-graph embeddings. Each category better fits the needs of different applications and problems. In this paper, we focus on the information that is used in addition to the plain structure by some embedding approaches. Especially, we study the impact of the node labels that are used by Graph2vec, a whole-graph embedding method. We study the effectiveness of such additional information in the context of online abuse detection.

The automated detection of abusive content on social media websites faces a variety of challenges including imbalanced training sets, the identification of an appropriate feature representation and the selection of optimal classifiers. Classifiers such as support vector machines (SVM), combined with bag of words or ngram feature representation, have traditionally dominated in text classification for decades. With the recent emergence of deep learning and word embeddings, an increasing number of researchers have started to focus on deep neural networks. In this paper, our aim is to explore cutting-edge techniques in automated abusive content detection. We use two deep learning approaches: convolutional neural networks (CNNs) and recurrent neural networks (RNNs). We apply these to 9 public datasets derived from various social media websites. Firstly, we show that word embeddings pre-trained on the same data source as the subsequent classification task improves the prediction accuracy of deep learning models. Secondly, we investigate the impact of different levels of training set imbalances on classifier types. In comparison to the traditional SVM classifier, we identify that although deep learning models can outperform the classification results of the traditional SVM classifier when the associated training dataset is seriously imbalanced, the performance of the SVM classifier can be dramatically improved through the use of oversampling, surpassing the deep learning models. Our work can inform researchers in selecting appropriate text classification strategies in the detection of abusive content, including scenarios where the training datasets suffer from class imbalance.

International audience; Online communities have gained considerable importance in recent years due to the increasing number of people connected to the Internet. Moderating user content in online communities is mainly performed manually, and reducing the workload through automatic methods is of great financial interest for community maintainers. Often, the industry uses basic approaches such as bad words filtering and regular expression matching to assist the moderators. In this article, we consider the task of automatically determining if a message is abusive. This task is complex since messages are written in a non-standardized way, including spelling errors, abbreviations, community-specific codes... First, we evaluate the system that we propose using standard features of online messages. Then, we evaluate the impact of the addition of pre-processing strategies, as well as original specific features developed for the community of an online in-browser strategy game. We finally propose to analyze the usefulness of this wide range of features using feature selection. This work can lead to two possible applications: 1) automatically flag potentially abusive messages to draw the moderator's attention on a narrow subset of messages ; and 2) fully automate the moderation process by deciding whether a message is abusive without any human intervention.

International audience; Moderating user content in online communities is mainly performed manually, and reducing the workload through automatic methods is of great interest. The industry mainly uses basic approaches such as bad words filtering. In this article, we consider the task of automatically determining whether a message is abusive or not. This task is complex, because messages are written in a non-standardized natural language. We propose an original automatic moderation method applied to French, which is based on both traditional tools and a newly proposed context-based feature relying on the modeling of user behavior when reacting to a message. The results obtained during this preliminary study show the potential of the proposed method, in a context of automatic processing or decision support.; La modération du contenu posté par les utilisateurs de communautés en ligne est ma-joritairement effectuée manuellement. De par la taille des données à traiter, les méthodes au-tomatiques ont un intérêt certain pour réduire la charge de travail. Actuellement, l'industrie utilise des approches basiques à base de recherche de mots, comme par exemple le filtrage des messages contenant certains mots interdits. Nous nous intéressons dans cet article à une tâche de classification permettant de déterminer si un message est abusif ou non. Ceci est com-plexe, car les messages sont écrits dans un langage naturel non standardisé. Nous proposons ici une approche originale de modération automatique appliquée au français, s'appuyant à la fois sur des outils classiques et un nouveau descripteur fondé sur la modélisation du comporte-ment utilisateur face à un message abusif. Les résultats obtenus lors de cette étude préliminaire montrent le potentiel de notre méthode, pour l'alerte automatique ou le support à la décision

El maltrato hacia las personas mayores es un fenómeno que se viene produciendo desde hace varias décadas, sin embargo, por diferentes razones como es el silencio tanto de la víctima como del cuidador, la falta de información de los profesionales o de la sociedad en general, comenzó a hacerse visible mucho más tarde. Actualmente, es un problema latente que se incrementa a medida que aumenta el envejecimiento de la población, afectando a la Salud Pública y Social, requiriendo por ello de una definición consensuada y una actuación conjunta por parte de los diferentes profesionales a los que compete. Un conocimiento general por parte de la población y de los profesionales implicados, así como el desarrollo de campañas de promoción y prevención, protocolos comunes de actuación o un incremento en la investigación, facilitaría la respuesta ante este fenómeno. Por ello, el principal objetivo, es evaluar la situación actual así como conocer los principales métodos de detección y abordaje. Destacando el papel conjunto de la enfermería junto con los Servicios Sociales en la detección, prevención y abordaje del mismo. Although elder mistreatment has occurred for several decades, it has not been recognized until quite recently. Patients and caregivers’ silence as well as professionanls’ lack of information’ and society’s underrecognition of the problem have kept this issue unexplored. Currently, it is a latent problem which increases at the same time that population grows old, affecting Public and Social Health. Therefore, it is necessary all professionals involved come to a definition of the problem by consensus and develop a joint intervention. Different actions would ease a consistent response to the problem: greater knowledge of this phenomenon on the part of both professionals and other members of society, the development of promotion and prevention campaigns, as well as common intervention protocols and an investment in the investigation of this issue. For this reason, this dissertation aims to assess the current situation as well as to define the main methods of detection and intervention. At the same time it will always highlight the importance of a collaborative role of Nursing and Social Services in the detection, prevention and intervention in cases of elder mistreatment. Grado en Enfermería

International audience; While online communities have become increasingly important over the years, the moderation of user-generated content is still performed mostly manually. Automating this task is an important step in reducing the financial cost associated with moderation, but the majority of automated approaches strictly based on message content are highly vulnerable to intentional obfuscation. In this paper, we discuss methods for extracting conversational networks based on raw multi-participant chat logs, and we study the contribution of graph features to a classification system that aims to determine if a given message is abusive. The conversational graph-based system yields unexpectedly high performance , with results comparable to those previously obtained with a content-based approach.

While online communities have become increasingly important over the years, the moderation of user-generated content is still performed mostly manually. Automating this task is an important step in reducing the financial cost associated with moderation, but the majority of automated approaches strictly based on message content are highly vulnerable to intentional obfuscation. In this paper, we discuss methods for extracting conversational networks based on raw multi-participant chat logs, and we study the contribution of graph features to a classification system that aims to determine if a given message is abusive. The conversational graph-based system yields unexpectedly high performance, with results comparable to those previously obtained with a content-based approach. MOTS-CLÉS : Catégorisation de texte, Détection d'abus, Communautés en ligne, Modération, Le nombre et la taille des communautés en ligne ne cessent de s'accroître depuis l'ap-parition du Web. Cependant, la modération du contenu produit par leurs utilisateurs est toujours essentiellement réalisée manuellement. L'automatisation de cette tâche permettrait de réduire son coût financier, mais la majorité des approches utilisées en production s'appuient unique-ment sur le contenu des messages et apparaissent très sensibles aux méthodes d'obfuscation intentionnelles. Dans cet article, nous proposons une méthode pour l'extraction de réseaux conversationnels à partir des logs de conversations (chat), et nous étudions le pouvoir discri-minant d'un ensemble de mesures topologiques pour un problème de classification consistant à identifier les messages abusifs. Notre approche atteint un niveau de performance inattendu, comparable à celui obtenu précédemment sur les mêmes données avec une approche unique-ment basée sur le contenu.

International audience; Future online social networks need to not only protect sensitive data of their users, but also protect them from abusive behavior coming from malicious participants in the network. We investigate the use of supervised learning techniques to detect abusive behavior and describe privacy-preserving protocols to compute the feature set required by abuse classification algorithms in a secure and privacy-preserving way. While our method is not yet fully resilient against a strong adaptive adversary, our evaluation suggests that it will be useful to detect abusive behavior with a minimal impact on privacy.

International audience; In this position paper we present the challenge of detecting abuse in a modern Online Social Network (OSN) while balancing data utility and privacy, with the goal of limiting the amount of user sensitive information processed during data collection, extraction and analysis. While we are working with public domain data available in a contemporary OSN, our goal is to design a thorough method for future alternative OSN designs that both protect user's sensitive information and discourage abuse. In this summary, we present initial results for detecting abusive behavior on Twitter. We plan to further investigate the impact of reducing input metadata on the quality of the abuse detection. In addition, we will consider defeating Byzantine behavior by opponents in the system.

This paper explores the question of whether school drug testing is an effective solution to tackle adolescent substance abuse problems. Research studies in major academic databases and Internet websites are reviewed. Several observations are highlighted from the review: (1) there are few research studies in this area, particularly in different Chinese contexts; (2) the quality of the existing studies was generally low; and (3) research findings supporting the effectiveness of school drug testing were mixed. Methodological issues underlying quantitative and qualitative evaluation studies of the effectiveness of school drug testing are also discussed.

© 2014 Harries et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. This article has been made available through the Brunel Open Access Publishing Fund. Background - Health and social care professionals are well positioned to identify and intervene in cases of elder financial abuse. An evidence-based educational intervention was developed to advance practitioners’ decision-making in this domain. The objective was to test the effectiveness of a decision-training educational intervention on novices’ ability to detect elder financial abuse. The research was funded by an E.S.R.C. grant reference RES-189-25-0334. Methods - A parallel-group, randomised controlled trial was conducted using a judgement analysis approach. Each participant used the World Wide Web to judge case sets at pre-test and post-test. The intervention group was provided with training after pre-test testing, whereas the control group were purely given instructions to continue with the task. 154 pre-registration health and social care practitioners were randomly allocated to intervention (n78) or control (n76). The intervention comprised of written and graphical descriptions of an expert consensus standard explaining how case information should be used to identify elder financial abuse. Participants’ ratings of certainty of abuse occurring (detection) were correlated with the experts’ ratings of the same cases at both stages of testing. Results - At pre-test, no differences were found between control and intervention on rating capacity. Comparison of mean scores for the control and intervention group at pre-test compared to immediate post-test, showed a statistically significant result. The intervention was shown to have had a positive moderate effect; at immediate post-test, the intervention group’s ratings had become more similar to those of the experts, whereas the control’s capacity did not improve. The results of this study indicate that the decision-training intervention had a positive effect on detection ability. Conclusions - This freely available, web-based decision-training aid is an effective evidence-based educational resource. Health and social care professionals can use the resource to enhance their ability to detect elder financial abuse. It has been embedded in a web resource at http://www.elderfinancialabuse.co.uk. ESRC

El maltrato en el anciano emerge como problema social hace pocos años y no porque antes no existiera, sino por el contrario siempre ha existido como un problema oculto, sin tener la verdadera importancia en la sociedad, debido al aumento de las expectativas de vida de los ancianos, el número de casos de maltrato se está incrementando y hay que tener en cuenta el impacto de este abuso sobre la salud del anciano. En el presente trabajo abordaremos conceptos básicos sobre el maltrato en el anciano y el papel de enfermería en el proceso de detección, para la detección del maltrato es importante contar con un instrumento que nos permita hacer una valoración general, rápida y concisa, donde se pueda recoger todos los datos, es este uno de los objetivos del presente trabajo, el hecho de tener un instrumento de valoración, permitirá realizar una evaluación adecuada de cada situación, con lo que conseguiremos detectar situaciones potenciales de maltrato. Grado en Enfermería

El objetivo de nuestra investigación ha sido evaluar la actitud afectiva, cognoscitiva y conductual del personal sanitario de las urgencias y emergencias extra hospitalarias de la provincia de Huelva y Sevilla. Se parte del papel tan importante que pueden jugar los servicios sanitarios para ayudar a las mujeres que sufren maltrato. Los servicios de urgencias y emergencias suponen una vía de fácil acceso y uno de los principales lugares de protección social al que acuden las víctimas, ya que existe una atención continuada durante las 24 horas del día, existe libertad de acceso a ellos o con una simple llamada telefónica por las diferentes líneas de atención de urgencias y/o emergencias sanitarias se desplazan hasta el lugar que la paciente solicite. A pesar de situarse en un lugar privilegiado para la adecuada detección y canalización de los posibles casos de maltrato que atienden existen numerosas dificultades y barreras entre las que destaca la escasa formación específica en violencia de género de este personal lo que lleva a desconocer en muchos casos cómo actuar en caso de maltrato. Los profesionales sanitarios además de tener capacidad de detección de casos de violencia de género tienen capacidad de intervenir, tener acceso a información específica y mediar para actuar y sobre todo a realizar una adecuada coordinación con los otros niveles asistenciales (atención primaria y servicios sociales). Para una eficaz coordinación es necesaria la existencia de equipos profesionales (psicólogos, médicos, enfermeros/as y trabajadores/as sociales) que se dediquen a ello sin interferir en las tareas asistenciales del servicio. Para cumplir nuestro objetivo se realizó un estudio trasversal mediante cuestionario de auto aplicación que se les proporciono a los sujetos. En una primera fase preparatoria por un lado se realizó una búsqueda bibliográfica sobre nuestro tema de estudio aclarando algunos conceptos fundamentales para entender el objeto de estudio, terminando con la construcción del marco teórico conceptual de la investigación. La segunda parte de esta primera fase fue el diseño de la investigación y la construcción y validación del cuestionario. Una vez elaborado el cuestionario se tuvo por un lado en cuenta los cuestionarios existentes en estudios similares y las aportaciones de diferentes expertos en la materia. Una vez pasada la prueba de jueces se modificó teniendo en cuenta las recomendaciones, se administró a 50 profesionales sanitario de los servicios de urgencias y emergencias de Andalucía. El análisis de los resultados mostró una buena fiabilidad y validez del instrumento. Una vez validada el cuestionario se administró al personal objeto de nuestro estudio obteniendo un nivel de respuesta del 70%. Los datos obtenidos con los cuestionarios se sometieron análisis empleando el software estadístico SPSS 17.0 que nos permitió además de operaciones básicas acceder a operaciones multivalente. Los resultados obtenidos muestran como el personal sanitario encuestado posee un nivel de conocimiento medio en materia de violencia de género y una actitud favorable en la detección y canalización de los casos de violencia de género. Aunque se destaca el papel relevante que dicho personal tiene en las actuaciones frente a la violencia de género la formación específica en dichos temas sigue siendo insuficiente, es necesario seguir potenciar la capacitación de este personal tanto desde los nuevos estudios de grado como desde el postgrado. Debe ser capaz de reconocer la magnitud y las repercusiones del problema, y la oportunidad que tienen de apoyar a las mujeres maltratadas en búsqueda de soluciones a su problema., The aim of this study has been the evaluation of affective cognitive and behavioral attitude of emergency and hospital emergency health personnel in the province of Huelva and Sevilla. We started from the important role that health services can play in order to help women suffering spousal abuse Emergency Services are an easy access route and one of the main social protection organizations for victims. They attend a call during 24 hours a day, with free access or moving their personnel to the place where the patient requests them as response to a simple phone call by the different lines of emergency care and / or health emergencies. Although they are situated in a prime location for proper detection and referral of potential cases of abuse, in many cases there are many difficulties and barriers in health staff to ignore leading how to act in spousal abuse. The most important of them is the Jack of specific training in domestic violence. Health professionals have ability to detect cases of domestic violence. In addition, they are able to intervene, to access specific information, to mediate to act and especially to make proper coordination with other levels of care (primary care and social services). It is necessary, for effective coordination, to have professional teams (psychologists, doctors, nurses, and social workers) dedicate to it, without interfering with the health care service. To achieve our goal we conducted a cross-sectional study using self-application questionnaire that was provided to subjects. In the preparatory phase, firstly, we made a literature search on our subject clarifying sorne basic concepts to understand the object of study, ending with the construction of the conceptual framework of the investigation. The second part of this phase was the research design and the construction and validation of the questionnaire. Once the questionnaire was developed we took into account existing questionnaires in similar studies and input from various experts. After the judges' test was passed the questionnaire was modified taking into account their recommendations. It was filled out by 50 health professionals from Emergency Services in Andalucía. The analysis of the results showed good reliability and validity of the instrument. Finally, the validated questionnaire was administered to staff subject of our study obtained a response, rate of 70%. Data from the questionnaires underwent analysis using SPSS 17.0 statistical software. It allowed us access to basic and multivalent operations. Results showed how the respondent medical staff had a medium level of knowledge about gender violence and a positive attitude in the detection and referral of cases of gender violence. In conclusion, although highlighting the important role that personnel has in the proceedings against gender violence, specific training in such subjects is still insufficient. It is necessary enhancing on training of such personnel from both the new grade studies and from the postgraduate degree. They must be able to recognize the magnitude and impact of the problem and the opportunity they have to support battered women seeking solutions to their problem.

[no abstract]