The rapid adoption of social media platforms by the general public has enabled the unprecedented spread of harmful content, often under the guise of anonymity, causing significant risks to individuals and communities, especially to vulnerable groups such as children, women, immigrants, etc. To stop the spread of such content and protect these groups, network immunization approaches from the epidemics domain have been employed. These immunization strategies use global spectral features of the graph, which struggle to scale to large-scale graphs that model real-world social media interactions and often fail to respond quickly to dynamic and community-localized outbreaks. To address this shortcoming, in this work, we introduce MeshShield , a novel immunization strategy that leverages the sparsity and community structure of social networks to enable efficient mitigation of harmful content by directly targeting infectious nodes and their neighbors. MeshShield applies a divide-et-impera approach to dynamically extract and analyze localized subgraphs centered around detected infectious sources, i.e., communities. Once these subgraphs are extracted, MeshShield employs a novel degree-weighted vulnerability measure to calculate which nodes within the community to immunize first. MeshShield analyzes all the subgraphs using a highly parallelized, shared-memory approach, thus reducing computational overhead and achieving near real-time performance even on large-scale real-world networks. Compared to state-of-the-art algorithms, such as DAVA and CONTAIN, the experimental results on real-world Twitter data demonstrate that our proposed strategy manages to immunize the network much faster while scaling with the graph’s size.

In today’s digital age, fake news has become a major problem with serious consequences, ranging from social unrest to political upheaval. New methods for detecting and mitigating fake news are required to address this issue. In this work, we propose incorporating contextual and network-aware features into the detection process. This involves analyzing not only the content of a news article but also the context in which it was shared and the network of users who shared it, i.e., the information diffusion. Thus, we propose GETAE, G raph Information E nhanced Deep Neural Ne T work Ensemble A rchitectur E for Fake News Detection, a novel ensemble architecture that uses textual content together with the social interactions to improve fake news detection. GETAE contains two Branches: the Text Branch and the Propagation Branch. The Text Branch combines Word and Transformer embeddings with a Deep Neural Network architecture based on feed-forward and bidirectional Recurrent Neural Networks ( [Bi]RNN ) to capture contextual features and generate a Text Content Embedding. This integrated approach allows for a more comprehensive understanding of the textual information. The Propagation Branch considers the information propagation within the graph network and proposes a Deep Learning architecture that employs Node Embeddings to create novel Propagation Embedding. GETAE’s Ensemble module combines the Text Content and Propagation Embeddings, to create a powerful and unique Propagation-Enhanced Content Embedding which is afterward used for classification. The experimental results obtained on two real-world publicly available datasets, i.e., Twitter15 and Twitter16, prove that this approach improves fake news detection and outperforms state-of-the-art models. • GETAE : graph-enhanced deep network for fake news detection using text and social data • Enhanced Text Content Embedding : combines complex lexical and syntactic features • New Propagation Embedding : captures information spread from a node to its network • Propagation-Enhanced Content Embedding : combines text, context, and propagation data • GETAE benchmark : validated via cross-validation, ablation, tuning on Twitter15 & 16

The increasing volume of online reviews has made the development of sentiment analysis models possible for determining customers’ opinions regarding different products and services. Until now, sentiment analysis has proven to be an effective tool for determining the overall polarity of reviews. To improve the granularity at the aspect level for a better understanding of the service or product, the task of aspect-based sentiment analysis aims to first identify aspects and then determine the user’s opinion about them. The complexity of this task lies in the fact that the same review can present multiple aspects, each with its own polarity. Current solutions have poor performance on such data. We address this problem by proposing ATESA-BÆRT, a heterogeneous ensemble learning model for Aspect-Based Sentiment Analysis. Firstly, we divide our problem into two sub-tasks, i.e., Aspect Term Extraction and Aspect Term Sentiment Analysis. Secondly, we use the argmax multi-class classification on six transformers-based learners for each sub-task. The proposed ensemble integrates representations from pre-trained and fine-tuned BERT and BART models in order to capture diverse linguistic features. By combining these with Linear , BiLSTM , and CNN-BiLSTM models, we aim to achieve enhanced performance on the aspect-based sentiment analysis task. Initial experiments on two publicly available real-world English online review datasets prove that ATESA-BÆRT outperforms current state-of-the-art solutions while tackling the sentiment analysis many aspects problem.

As global digitization continues to grow, technology becomes more affordable and easier to use, and social media platforms thrive, becoming the new means of spreading information and news. Communities are built around sharing and discussing current events. Within these communities, users are enabled to share their opinions about each event. Using Sentiment Analysis to understand the polarity of each message belonging to an event, as well as the entire event, can help to better understand the general and individual feelings of significant trends and the dynamics on online social networks. In this context, we propose a new ensemble architecture, EDSA-Ensemble (Event Detection Sentiment Analysis Ensemble), that uses Event Detection and Sentiment Analysis to improve the detection of the polarity for current events from Social Media. For Event Detection, we use techniques based on Information Diffusion taking into account both the time span and the topics. To detect the polarity of each event, we preprocess the text and employ several Machine and Deep Learning models to create an ensemble model. The preprocessing step includes several word representation models: raw frequency,$TFIDF$, Word2Vec, and Transformers. The proposed EDSA-Ensemble architecture improves the event sentiment classification over the individual Machine and Deep Learning models.

Understanding the relation between the meanings of words is an important part of comprehending natural language. Prior work has either focused on analysing lexical semantic relations in word embeddings or probing pretrained language models (PLMs), with some exceptions. Given the rarity of highly multilingual benchmarks, it is unclear to what extent PLMs capture relational knowledge and are able to transfer it across languages. To start addressing this question, we propose MultiLexBATS, a multilingual parallel dataset of lexical semantic relations adapted from BATS in 15 languages including low-resource languages, such as Bambara, Lithuanian, and Albanian. As experiment on cross-lingual transfer of relational knowledge, we test the PLMs’ ability to (1) capture analogies across languages, and (2) predict translation targets. We find considerable differences across relation types and languages with a clear preference for hypernymy and antonymy as well as romance languages.

With advances in the field of Linked (Open) Data (LOD), language data on the LOD cloud has grown in number, size, and variety. With an increased volume and variety of language data, optimizations of methods for distributing, storing, and querying these data become more central. To this end, this position paper investigates use cases at the intersection of LLOD and Big Data, existing approaches to utilizing Big Data techniques within the context of linked data, and discusses the challenges and benefits of this union.

With the emergence of social networks, online platforms dedicated to different use cases, and sensor networks, the emergence of large-scale graph community detection has become a steady field of research with real-world applications. Community detection algorithms have numerous practical applications, particularly due to their scalability with data size. Nonetheless, a notable drawback of community detection algorithms is their computational intensity [2], resulting in decreasing performance as data size increases. For this purpose, new frameworks that employ distributed systems such as Apache Hadoop and Apache Spark which can seamlessly handle large-scale graphs must be developed. In this paper, we propose a novel framework for community detection algorithms, i.e., K-Cliques, Louvain, and Fast Greedy, developed using Apache Spark GraphFrames. We test their performance and scalability on two real-world datasets. The experimental results prove the feasibility of developing graph mining algorithms using Apache Spark GraphFrames.

The growing popularity of social media platforms has simplified the creation and distribution of news articles but also creates a conduit for spreading fake news. In consequence, the need arises for effective context-aware fake news detection mechanisms, where the contextual information can be built either from the textual content of posts or from available social data (e.g., information about the users, reactions to posts, or the social network). In this paper, we propose DANES, a Deep Neural Network Ensemble Architecture for Social and Textual Context-aware Fake News Detection. DANES comprises a Text Branch for a textual content-based context and a Social Branch for the social context. These two branches are used to create a novel Network Embedding. Preliminary ablation results on 3 real-world datasets, i.e., BuzzFace, Twitter15, and Twitter16, are promising, with an accuracy that outperforms state-of-the-art solutions when employing both social and textual content features. In the present setting, with so much manipulation on social media platforms, our solution can enhance fake news identification even with limited training data.

Real-time social media data can provide useful information on evolving hazards. Alongside traditional methods of disaster detection, the integration of social media data can considerably enhance disaster management. In this paper, we investigate the problem of detecting geolocation-content communities on Twitter and propose a novel distributed system that provides in near real-time information on hazard-related events and their evolution. We show that content-based community analysis can lead to better and faster dissemination of hazard-related reports than using only traditional methods, such as satellite or airborne sensing platforms. Our distributed disaster reporting system analyzes the social relationship among worldwide geolocated tweets and applies topic modeling to group tweets by topics. Considering for each tweet the following information: user, timestamp, geolocation, retweets, and replies, we create a publisher-subscriber distribution model for topics. We use content similarity and the proximity of nodes to create a new model for geolocation-content based communities. Users can subscribe to different topics in specific geographical areas or worldwide and receive real-time reports regarding these topics. As misinformation can lead to increased damage if propagated in hazards-related tweets, we propose a new deep learning model to detect fake news. The misinformed tweets are then removed from display. We also show empirically the scalability capabilities of the proposed system.

Automatic Term Recognition is used to extract domain-specific terms that belong to a given domain. In order to be accurate, these corpus and language-dependent methods require large volumes of textual data that need to be processed to extract candidate terms that are afterward scored according to a given metric. To improve text preprocessing and candidate terms extraction and scoring, we propose a distributed Spark-based architecture to automatically extract domain-specific terms. The main contributions are as follows: (1) propose a novel distributed automatic domain-specific multi-word term recognition architecture built on top of the Spark ecosystem; (2) perform an in-depth analysis of our architecture in terms of accuracy and scalability; (3) design an easy-to-integrate Python implementation that enables the use of Big Data processing in fields such as Computational Linguistics and Natural Language Processing. We prove empirically the feasibility of our architecture by performing experiments on two real-world datasets.

OntoLex-Lemon has become a de facto standard for lexical resources in the web of data. This paper provides the first overall description of the emerging OntoLex module for Frequency, Attestations, and Corpus-Based Information (OntoLex-FrAC) that is intended to complement OntoLex-Lemon with the necessary vocabulary to represent major types of information found in or automatically derived from corpora, for applications in both language technology and the language sciences.

In this paper, we provide an overview of current technologies for cross-lingual link discovery, and we discuss challenges, experiences and prospects of their application to under-resourced languages. We rst introduce the goals of cross-lingual linking and associated technologies, and in particular, the role that the Linked Data paradigm (Bizer et al., 2011) applied to language data can play in this context. We de ne under-resourced languages with a speci c focus on languages actively used on the internet, i.e., languages with a digitally versatile speaker community, but limited support in terms of language technology. We argue that languages for which considerable amounts of textual data and (at least) a bilingual word list are available, techniques for cross-lingual linking can be readily applied, and that these enable the implementation of downstream applications for under-resourced languages via the localisation and adaptation of existing technologies and resources.

Discourse markers carry information about the discourse structure and organization, and also signal local dependencies or epistemological stance of speaker. They provide instructions on how to interpret the discourse, and their study is paramount to understand the mechanism underlying discourse organization. This paper presents a new language resource, an ISO-based annotated multilingual parallel corpus for discourse markers. The corpus comprises nine languages, Bulgarian, Lithuanian, German, European Portuguese, Hebrew, Romanian, Polish, and Macedonian, with English as a pivot language. In order to represent the meaning of the discourse markers, we propose an annotation scheme of discourse relations from ISO 24617-8 with a plug-in to ISO 24617-2 for communicative functions. We describe an experiment in which we applied the annotation scheme to assess its validity. The results reveal that, although some extensions are required to cover all the multilingual data, it provides a proper representation of discourse markers value. Additionally, we report some relevant contrastive phenomena concerning discourse markers interpretation and role in discourse. This first step will allow us to develop deep learning methods to identify and extract discourse relations and communicative functions, and to represent that information as Linguistic Linked Open Data (LLOD).

Developing artificial learning systems that can understand and generate natural language has been one of the long-standing goals of artificial intelligence. Recent decades have witnessed an impressive progress on both of these problems, giving rise to a new family of approaches. Especially, the advances in deep learning over the past couple of years have led to neural approaches to natural language generation (NLG). These methods combine generative language learning techniques with neural-networks based frameworks. With a wide range of applications in natural language processing, neural NLG (NNLG) is a new and fast growing field of research. In this state-of-the-art report, we investigate the recent developments and applications of NNLG in its full extent from a multidimensional view, covering critical perspectives such as multimodality, multilinguality, controllability and learning strategies. We summarize the fundamental building blocks of NNLG approaches from these aspects and provide detailed reviews of commonly used preprocessing steps and basic neural architectures. This report also focuses on the seminal applications of these NNLG models such as machine translation, description generation, automatic speech recognition, abstractive summarization, text simplification, question answering and generation, and dialogue generation. Finally, we conclude with a thorough discussion of the described frameworks by pointing out some open research directions.

With the rise of social media users and the general shift of communication from traditional media to online platforms, the spread of harmful content (e.g., hate speech, misinformation, fake news) has been exacerbated. Harmful content in the form of hate speech causes a person distress or harm, having a negative impact on the individual mental health, with even more detrimental effects on the psychology of children and teenagers. In this paper, we propose an end-to-end solution with real-time capabilities to detect harmful content in real-time and mitigate its spread over the network. Our main contribution is Sparse Shield, a novel method that out-scales existing state-of-the-art methods for network immunization. We also propose a novel architecture for harmful speech mitigation that maximizes the impact of immunization. Our solution aims to identify a set of users for which to move harmful content at the bottom of the user feed, rather than censoring users. By immunizing certain network nodes in this manner, we minimize the negative impact on the network and minimize the interference with and limitation of individual freedoms: the information is not hidden but rather not as easy to reach without an explicit search. Our analysis is based on graphs built on real-world data collected from Twitter; these graphs reflect real user behavior. We perform extensive scalability experiments to prove the superiority of our method over existing state-of-the-art network immunization techniques. We also perform extensive experiments to showcase that Sparse Shield outperforms existing techniques on the task of harmful speech mitigation on a real-world dataset.

Personalized social media offer communication opportunities that mass media could not afford, yet also raise novel challenges. A prime challenge arising from this shift in digital communication is to detect topics and events of interest. In this paper, we propose and deploy a novel Deep Learning architecture that predicts if a news topic becomes viral by analyzing social media diffusion and audience interest in current news events. The proposed solution: (i) analyzes news articles, (ii) extracts associated topics and events, (iii) matches the topics and events to filter and extract developing topics, (iv) extracts current events from Twitter and matches them to the filtered news topics, and (v) predicts audience interest in news topics using Twitter likes and retweets. We employ several feature engineering techniques to improve prediction by integrating user metadata into the training set. In our experiments, we correlate two datasets collected over several months in the same time period. The first dataset contains news articles collected from different news venues, while the second one contains tweets regarding the news. The experimental results from our real-world deployment prove that the proposed system achieves high accuracy when integrating influencers metadata and the day of the week. Thus, proving that the news topics virality prediction is improved under the assumptions that spreaders and the day of the week play a huge role in information diffusion.

Apache Spark is a distributed computing framework for fast in-memory data analysis, Machine Learning jobs, and SQL queries that employs Resilient Distributed Dataset for distributing data and Directed Acyclic Graph for scheduling computations. Currently, Spark provides two scheduling policies for tasks pending execution: a First In First Out policy and a FAIR policy, providing no support for deadline-based real-time scheduling. In this paper, we present a new system designed to accept deadlines for heterogeneous Spark Jobs and perform a real-time scheduling policy based on Earliest Deadline First (EDF). To showcase the efficiency of our scheduling policy, we compare and analyze the performance of our solution with the current Spark execution policies in terms of job lateness. We empirically prove that our real-time policy provides much lower lateness given suitable constraints.

The paper proposes an interdisciplinary approach including methods from disciplines such as history of concepts, linguistics, natural language processing (NLP) and Semantic Web, to create a comparative framework for detecting semantic change in multilingual historical corpora and generating diachronic ontologies as linguistic linked open data (LLOD). Initiated as a use case (UC4.2.1) within the COST Action Nexus Linguarum, European network for Web-centred linguistic data science, the study will explore emerging trends in knowledge extraction, analysis and representation from linguistic data science, and apply the devised methodology to datasets in the humanities to trace the evolution of concepts from the domain of socio-cultural transformation. The paper will describe the main elements of the methodological framework and preliminary planning of the intended workflow.

Radu Cristian Alexandru Iacob, Florin Brad, Elena-Simona Apostol, Ciprian-Octavian Truică, Ionel Alexandru Hosu, Traian Rebedea. Proceedings of the 28th International Conference on Computational Linguistics. 2020.