Search

Uma das estratégias para combater o crescente problema das fake news é capacitar pessoas para identificar esse tipo de notícia. Embora existam iniciativas em que tal capacitação é apoiada por jogos educacionais digitais (JED), os JED utilizados não dispõem de notícias escritas em Língua Portuguesa. Para suprir esta lacuna, o artigo apresenta o JEDi, um JED voltado à capacitação de estudantes na identificação de notícias falsas, divulgadas de forma intencional e no idioma português. O JEDi se desenrola em um tabuleiro a ser percorrido pelos jogadores na medida em que esses conseguem discernir entre notícias verdadeiras e falsas. Vence a partida, o jogador que alcançar o final do tabuleiro primeiro. A ideia é que, na medida em que joguem diversas partidas, os jogadores desenvolvam a capacidade de reconhecer notícias falsas. Ao persistir os resultados individuais dos jogadores, o JEDi permite analisar, comtécnicas de mineração de dados, o desempenho longitudinal de cada jogador e, portanto, sua efetividade como jogo na capacitação para reconhecer fake news. O artigo relata a aplicação do JEDi em um estudo de caso com alunos do Ensino Médio e outro com alunos da Educação Superior. Resultados quantitativos e qualitativos obtidos nos dois estudos apontam para a efetividade do JEDi como instrumento de capacitação na identificação de fake news.

A identificação de pessoas suspeitas de crimes em redes sociais (e.g. pedofilia, terrorismo etc.) tem tido destaque nos últimos anos. Contudo, na literatura, as interações geradas, a partir de conteúdos textuais postados nessas redes, nem sempre são consideradas. Desse modo, o presente trabalho apresenta o algoritmo, denominado TROY, capaz de explicitar essas interações, bem como seus impactos, a fim de apoiar a identificação de suspeitos. Além disso, perante as dificuldades em se obter dados em português para experimentos, este artigo apresenta uma nova forma de construção de conjuntos de dados para experimentos, utilizando a tarefa de predição de links. Os resultados obtidos, por meio de experimentos, demonstram uma melhora na identificação de suspeitos.

Internet of Things (IoT) systems usually involve interconnected, low processing capacity, and low memory sensor nodes (devices) that collect data in several sorts of applications that interconnect people and things. In this scenario, mining tasks, such as clustering, have been commonly deployed to detect behavioral patterns from the collected data. The centralized clustering of IoT data demands high network traffic to transmit the data from the devices to a central node, where a clustering algorithm must be applied. This approach does not scale as the number of devices increases, and the amount of data grows. However, distributing the clustering process through the devices may not be a feasible approach as well, since the devices are usually simple and may not have the ability to execute complex procedures. This work proposes a centralized IoT data clustering method that demands reduced network traffic and low processing power in the devices. The proposed method uses a data grid to summarize the information at the devices before transmitting it to the central node, reducing network traffic. After the data transfer, the proposed method applies a clustering algorithm that was developed to process data in the summarized representation. Tests with seven datasets provided experimental evidence that the proposed method reduces network traffic and produces results comparable to the ones generated by DBSCAN and HDBSCAN, two robust centralized clustering algorithms.

A Web de Dados é uma relevante e crescente fonte de dados que contém informações distribuídas em diferentes datasets interligados. A maioria dos algoritmos de mineração de dados foi projetada para analisar um único dataset por vez, e, consequentemente, não consegue explorar as conexões entre datasets da Web de Dados. Para suprir essa lacuna, este trabalho propõe o MRAR+, um algoritmo de mineração de grafos que busca por regras de associação multirrelação a fim de identificar conhecimentos novos que envolvam recursos de múltiplos datasets da Web de Dados. O MRAR+ foi aplicado com sucesso em dois experimentos e produziu regras novas e úteis para os usuários, ilustrando a sua viabilidade para minerar diferentes datasets interligados na Web de Dados.

The dissemination of news through digital media has amplified Fake News proliferation. In the face of this scenario, sentiment-based methods have presented promising results in Fake News detection. Although sentiment-based methods can extract sentiment (i.e., polarity and/or emotion) from either texts or images available in news, the ones applied to Portuguese-written news have considered sentiment exclusively extracted from texts. Thus, this study proposes a multimodal method that, besides the polarity and emotions extracted from texts, also considers sentiment extracted from news' images in order to detect Fake News written in Portuguese. The proposed method showed promising results in experimental data, overcoming the baseline methods in 8 p.p.

The identification of crime suspects on social networks (e.g. pedophilia, terrorism, etc.) has been one of the most relevant topics in social network analysis. The vast majority of methods are based on the extraction of textual content, senders and receivers from the network, but do not take into account the interactions that occur inside the textual content. The present work raises the hypothesis that these interactions, if taken into consideration, can lead to better results when it comes to identifying crime suspects. To validate this hypothesis, an algorithm named TROY was developed to define and represent the interactions that occur in Youtube. Then, it is proposed as the first stage of the INSPECTION method, which analyzes the textual content exchanged on a social network, in order to identify crime suspects. This method is based on the use of a controlled vocabulary with terms categorized according to a certain domain (for example, pedophilia, cyberbullying, terrorism, etc.). Experiments on the pedophilia domain were carried out applying the INSPECTION method to Youtube. The first stage of the method may use any algorithm that is able to extract social interactions. In this case, we used TROY and another algorithm called CRAWLER, which does not take into account the previously mentioned interactions. The results showed that the TROY algorithm obtained better results than the CRAWLER algorithm, validating the hypothesis raised.

Social Commerce has risen and evolved in the last years due to changes either in e-commerce or social networks applications. On top of that, the number of online ads and transactions in Social Commerce has grown. This environment is attractive to either good users and bad users. The bad users cause harm to their victims by making them lose money or suffer psychological damage. Since the volume of transactions is high and the fraud occurrence is low, the manual detection is highly inefficient (too much resource required for low detection) and unscalable. The existing solutions for automatic fraud detection in Social Commerce are based on structured information available in ads such as price, product type, brand, new/used, among others. However, such solutions ignore possible fraud signs from the ads’ images that exhibit the products sold. Therefore, this work aims to evaluate if combining structured information and images available in the ads provides more effective models than the ones considering only structured information. To this end, it proposes FDSC, a method that combines information obtained from ads’ images through deep learning with structured information available in the corresponding ads, in order to detect fraud in Social Commerce. Experimental evidence shows an incremental opportunity of 7% in F-score by the adoption of FDSC.

Community detection is an important network analysis task that has been studied by academy and industry for the last years. Community detection algorithms try to maximize the number of connections in each community and minimize the number of connections between different communities. Some of them consider not only the topological aspects of the networks but also try to explore existing information about the context of the application available in attributes of nodes and/or connections in order to find cohesive content communities. Those algorithmswere designed to run exclusively over homogeneous networks and cannot deal with heterogeneous structures. Nevertheless, typical real-world networks are heterogeneous. Thus, this article proposes ComDet, a community detection approach that fills this gap by taking into account topological and contextual information to detect communities in heterogeneous networks. The proposed approach uses data clustering as a pre-processing step for the community detection process in order to identify similar nodes that are directly or indirectly linked and organize them in cohesive and possibly overlapping communities. Experiments in three attributed heterogeneous networks show that ComDet leads to interesting partitions with cohesive content communities.

Background: In a phase 1b study, intravenous daratumumab plus pomalidomide and dexamethasone induced a very good partial response or better rate of 42% and was well tolerated in patients with heavily pretreated multiple myeloma. We aimed to evaluate whether daratumumab plus pomalidomide and dexamethasone would improve progression-free survival versus pomalidomide and dexamethasone alone in patients with previously treated multiple myeloma. Methods: In this ongoing, open-label, randomised, phase 3 trial (APOLLO) done at 48 academic centres and hospitals across 12 European countries, eligible patients were aged 18 years or older, had relapsed or refractory multiple myeloma with measurable disease, had an Eastern Cooperative Oncology Group performance status of 0–2, had at least one previous line of therapy, including lenalidomide and a proteasome inhibitor, had a partial response or better to one or more previous lines of antimyeloma therapy, and were refractory to lenalidomide if only one previous line of therapy was received. Patients were randomly assigned (1:1) by an interactive web-response system in a random block size of two or four to receive pomalidomide and dexamethasone alone or daratumumab plus pomalidomide and dexamethasone. Randomisation was stratified by number of previous lines of therapy and International Staging System disease stage. All patients received oral pomalidomide (4 mg, once daily on days 1–21) and oral dexamethasone (40 mg once daily on days 1, 8, 15, and 22; 20 mg for those aged 75 years or older) at each 28-day cycle. The daratumumab plus pomalidomide and dexamethasone group received daratumumab (1800 mg subcutaneously or 16 mg/kg intravenously) weekly during cycles 1 and 2, every 2 weeks during cycles 3–6, and every 4 weeks thereafter until disease progression or unacceptable toxicity. The primary endpoint was progression-free survival in the intention-to-treat population. Safety was analysed in all patients who received at least one dose of study medication. This trial is registered with ClinicalTrials.gov, NCT03180736. Findings: Between June 22, 2017, and June 13, 2019, 304 patients (median age 67 years [IQR 60–72]; 161 [53%] men and 143 [47%] women) were randomly assigned to the daratumumab plus pomalidomide and dexamethasone group (n=151) or the pomalidomide and dexamethasone group (n=153). At a median follow-up of 16·9 months (IQR 14·4–20·6), the daratumumab plus pomalidomide and dexamethasone group showed improved progression-free survival compared with the pomalidomide and dexamethasone group (median 12·4 months [95% CI 8·3–19·3] vs 6·9 months [5·5–9·3]; hazard ratio 0·63 [95% CI 0·47–0·85], two-sided p=0·0018). The most common grade 3 or 4 adverse events were neutropenia (101 [68%] of 149 patients in the daratumumab plus pomalidomide and dexamethasone group vs 76 [51%] of 150 patients in the pomalidomide and dexamethasone group), anaemia (25 [17%] vs 32 [21%]), and thrombocytopenia (26 [17%] vs 27 [18%]). Serious adverse events occurred in 75 (50%) of 149 patients in the daratumumab plus pomalidomide and dexamethasone group versus 59 (39%) of 150 patients in the pomalidomide and dexamethasone group; pneumonia (23 [15%] vs 12 [8%] patients) and lower respiratory tract infection (18 [12%] vs 14 [9%]) were most common. Treatment-emergent deaths were reported in 11 (7%) patients in the daratumumab plus pomalidomide and dexamethasone group versus 11 (7%) patients in the pomalidomide and dexamethasone group. Interpretation: Among patients with relapsed or refractory multiple myeloma, daratumumab plus pomalidomide and dexamethasone reduced the risk of disease progression or death versus pomalidomide and dexamethasone alone and could be considered a new treatment option in this setting. Funding: European Myeloma Network and Janssen Research and Development. © 2021 Elsevier Ltd

Due to easy access and low cost, social media online news consumption has increased significantly for the last decade. Despite their benefits, some social media allow anyone to post news with intense spreading power, which amplifies an old problem: the dissemination of Fake News. In the face of this scenario, several machine learning-based methods to automatically detect Fake News (MLFN) have been proposed. All of them require datasets to train and evaluate their detection models. Although recent MLFN were designed to consider data regarding the news propagation on social media, most of the few available datasets do not contain this kind of data. Hence, comparing the performances amid those recent MLFN and the others is restricted to a very limited number of datasets. Moreover, all existing datasets with propagation data do not contain news in Portuguese, which impairs the evaluation of the MLFN in this language. Thus, this work proposes FakeNewsSetGen, a process that builds Fake News datasets that contain news propagation data and support comparison amid the state-of-the-art MLFN. FakeNewsSetGen's software engineering process was guided to include all kind of data required by the existing MLFN. In order to illustrate FakeNewsSetGen's viability and adequacy, a case study was carried out. It encompassed the implementation of a FakeNewsSetGen prototype and the application of this prototype to create a dataset called FakeNewsSet, with news in Portuguese. Five MLFN with different kind of data requirements (two of them demanding news propagation data) were applied to FakeNewsSet and compared, demonstrating the potential use of both the proposed process and the created dataset.

In the last decades, the dissemination of News through digital media has increased the information accessibility previously offered by traditional channels. Despite their benefits, digital media have exacerbated an old problem: the spread of Fake News, (i.e., false News intentionally published). Faced with this scenario, the linguistic approaches to automatic Fake News detection use information that can be directly extracted from the News' text. Several methods based on these approaches use grammatical classification and sentiment analysis over News writing in Portuguese. However, as far as it was possible to observe in the related literature, these methods are limited to the identification of polarity of sentiment (i.e., positive, neutral or negative) existing in the text. Although polarity classification be an effective method for a wide range of natural language processing applications, it does not address language nuances (e.g., emotions such as anger, sadness, etc.) that can provide evidence that a text contains false information. Hence, this study proposes an extended method that, in addition to the grammatical classification and polarity based sentiment analysis, also uses the analysis of emotions to detect Fake News written in Portuguese. The extended method showed promising results in experimental data, obtaining accuracy greater than 92%. In average, the proposed method overcame polarity and gramatical classification based methods in 1.4 percentage points.

Identifying suspects of crimes on social networks is one of the most relevant tasks in the analysis of this type of network. Most of the computational methods focused on this task involve supervised machine learning, and, therefore, require previously labeled datasets that inform, among the registered people, messages and/or conversations, which ones are suspects. However, in practice, this type of information is not available, for several reasons, among which, it is rare or even protected by secrecy guaranteed by law. This limitation makes it very difficult to effectively use these methods in real situations. Hence, the present work raises the hypothesis that the use of a controlled vocabulary on the field of application can make it possible the identification of suspects in social networks, without the need for previously labeled datasets. In order to search for experimental evidence that points to the validity of the hypothesis raised, this article proposes a generic method that uses a controlled vocabulary with categorized terms, according to a certain domain (e.g., pedophilia, cyberbullying, terrorism, etc.), to analyze messages exchanged on social networks, in order to identify criminal suspects. The results obtained in a preliminary experiment in pedophilia domain showed signs of adequacy of the proposed method.

O problema das fake news tem preocupado os mais diversos segmentos sociais. Uma das estratégias para combater essas notícias é capacitar pessoas para identificá-las. Embora existam iniciativas em que tal capacitação é apoiada por jogos educacionais digitais (JED), os JED utilizados não dispõem de notícias escritas em Língua Portuguesa. Para suprir esta lacuna, o presente artigo propõe uma versão do "Jogo da Trilha'' que exercita a identificação de fake news escritas em Português, além de oferecer suporte para mineração de dados sobre o desempenho de seus jogadores. Um estudo de caso com quarenta e três alunos de ensino médio revelou evidências de que o JED proposto contribuiu para a capacitação discente na identificação de fake news.

The current pandemic of COVID-19 – Coronavirus Disease-19, caused by the pathogen SARS-CoV-2, has already killed thousands of people in the year of 2020. We did a review of literature to understand the potential use of AI to respond to COVID-19. The search strategy used five descriptors: (i) “Artificial Intelligence”, (ii) “Deep Learning”, (iii) “Machine Learning”, (iv) “COVID-19” and (v) “SARS-CoV-2”, which have been combined to search for articles in the PubMed and ResearchGate databases. The bibliographic search has been complemented with texts of previous knowledge of the authors. The subjects covered in the chosen bibliography have been organized into two main study categories: (i) AI for diagnosis; and (ii) AI for controlling dissemination. The final considerations point out the main complications for the proper use of AI and the possibilities of its application to support decision making of health professionals and at managerial levels of public health.

Link prediction is an online social network (OSN) analysis task whose objective is to identify pairs of non-connected nodes with a high probability of getting connected in the near future. Recently, proposed link prediction methods consider topological data from OSN past states (i.e., snapshots that depict the network structure at certain moments in the past). Although past states-based methods retrieve information that describes how the network’s topology was at the events of link emergence (i.e., moments when the existing edges were created), they do not take into account contextual data concerning those events. Hence, they take the chance to disregard information about the circumstances that may have influenced the appearance of old edges, and that could be useful to predict the creation of new ones. To remedy this issue, this work extends a past states-based method to retrieve both topological and contextual data from the events of edge emergence and combine them to predict links. The extended method presented promising results on experimental data. Overall, it overcame the original method in five different scenarios from five co-authorship OSN frequently used for link prediction method evaluation.

Understanding and characterizing the processes driving social interactions is one of the fundamental problems in social network analysis. In this context, link prediction aims to foretell whether two not linked nodes in a network will connect in the near future. Several studies proposed to solve link prediction compute compatibility degrees as link weights between connected nodes and, based on a weighted graph, apply weighted similarity functions to non-connected nodes in order to identify potential new links. The weighting criteria used by those studies were based exclusively on information about the existing topology (network structure). Nevertheless, such approach leads to poor incorporation of other aspects of the social networks, such as context (node and link attributes), and temporal information (chronological interaction data). Hence, in this paper, we propose three weighting criteria that combine contextual, temporal and topological information in order to improve results in link prediction. We evaluated the proposed weighting criteria with two popular weighted similarity functions (Adamic-Adar and Common Neighbors) in ten networks frequently used in experiments with link prediction. Results with the proposed criteria were statistically better than the ones obtained from the weighting criterion that is exclusively based on topological information.

The problem of automatic Fake News detection in digital media of news distribution (DMND - e.g., social networks, online newspaper, etc) has become even more relevant. Among the main detection approaches, the one based on crowd signals from DMND users has stood out by obtaining promising results. In essence, in order to classify a piece of news as fake or not fake, such approach explores the collective sense by combining opinions (signals, i.e., votes about the classification of some news) of a high number of users (crowd), considering the reputations of these users regarding their capacity of identifying Fake News. Although promising, the Crowd Signals approach has a significant limitation: it depends on the explicit user opinion (which is not always available) about the classification of the analyzed news. Such unavailability may be caused by the absence of a functionality in the DMND that collects user opinion about the news, or by the simple option of the users in not giving their opinion. Facing this limitation, the present work raises the hypothesis that it is possible to build models of Fake News detection with a performance comparable to the Crowd Signals based approach, avoiding the dependence on the explicit opinion of DMND users. To validate this hypothesis, the present work proposes HCS, an approach based on crowd signals that considers implicit user opinions instead of the explicit ones. The implicit opinions are inferred from the behavior of users concerning the dissemination of the news analyzed. Inspired in Meta-Learning, the HCS can also use the explicit opinions from machines (news classification models) to complement the implicit user opinions by means of hybrid Crowds. Experiments carried out in five datasets presented significant evidence that confirms the raised hypothesis. Even without considering DMND users’ explicit opinions, HCS was able to achieve results comparable to the ones produced by the Crowd Signals approach. Besides that, the results also revealed a performance improvement of HCS when the implicit opinions of the users were combined with the explicit opinions of machines.

For the last years, a considerable amount of attention has been devoted to the research about the link prediction (LP) problem in complex networks. This problem tries to predict the likelihood of an association between two not interconnected nodes in a network to appear in the future. One of the most important approaches to the LP problem is based on supervised machine learning (ML) techniques for classification. Although many works have presented promising results with this approach, choosing the set of features (variables) to train the classifiers is still a major challenge. In this article, we report on the effects of three different automatic variable selection strategies (Forward, Backward and Evolutionary) applied to the feature-based supervised learning approach in LP applications. The results of the experiments show that the use of these strategies does lead to better classification models than classifiers built with the complete set of variables. Such experiments were performed over three datasets (Microsoft Academic Network, Amazon and Flickr) that contained more than twenty different features each, including topological and domain-specific ones. We also describe the specification and implementation of the process used to support the experiments. It combines the use of the feature selection strategies, six different classification algorithms (SVM, K-NN, naive Bayes, CART, random forest and multilayer perceptron) and three evaluation metrics (Precision, F-Measure and Area Under the Curve). Moreover, this process includes a novel ML voting committee inspired approach that suggests sets of features to represent data in LP applications. It mines the log of the experiments in order to identify sets of features frequently selected to produce classification models with high performance. The experiments showed interesting correlations between frequently selected features and datasets.