Search

The advent of online platforms and services has revolutionized communication, enabling users to share opinions and ideas seamlessly. However, this convenience has also brought about a surge in offensive and harmful language across various communication mediums. In response, social platforms have turned to automated methods to identify offensive content. A critical research question emerges when investigating the role of specific text spans within comments in conveying offensive characteristics. This paper conducted a comprehensive investigation into detecting offensive text spans in Romanian language comments using Transformer encoders and Large Language Models (LLMs). We introduced an extensive dataset of 4800 Romanian comments annotated with offensive text spans. Moreover, we explored the impact of varying model sizes, architectures, and training data volumes on the performance of offensive text span detection, providing valuable insights for determining the optimal configuration. The results argue for the effectiveness of BERT pre-trained models for this span-detection task, showcasing their superior performance. We further investigated the impact of different sample-retrieval strategies for few-shot learning using LLMs based on vector text representations. The analysis highlights important insights and trade-offs in leveraging LLMs for offensive-language-detection tasks. [ABSTRACT FROM AUTHOR]

With the current shift in the mass media landscape from journalistic rigor to social media, personalized social media is becoming the new norm. Although the digitalization progress of the media brings many advantages, it also increases the risk of spreading disinformation, misinformation, and malformation through the use of fake news. The emergence of this harmful phenomenon has managed to polarize society and manipulate public opinion on particular topics, e.g., elections, vaccinations, etc. Such information propagated on social media can distort public perceptions and generate social unrest while lacking the rigor of traditional journalism. Natural Language Processing and Machine Learning techniques are essential for developing efficient tools that can detect fake news. Models that use the context of textual data are essential for resolving the fake news detection problem, as they manage to encode linguistic features within the vector representation of words. In this paper, we propose a new approach that uses document embeddings to build multiple models that accurately label news articles as reliable or fake. We also present a benchmark on different architectures that detect fake news using binary or multi-labeled classification. We evaluated the models on five large news corpora using accuracy, precision, and recall. We obtained better results than more complex state-of-the-art Deep Neural Network models. We observe that the most important factor for obtaining high accuracy is the document encoding, not the classification model's complexity. [ABSTRACT FROM AUTHOR]

With the increased popularity of social media platforms such as Twitter or Facebook, sentiment analysis (SA) over the microblogging content becomes of crucial importance. The literature reports good results for well-resourced languages such as English, Spanish or German, but open research space still exists for underrepresented languages such as Romanian, where there is a lack of public training datasets or pretrained word embeddings. The majority of research on Romanian SA tackles the issue in a binary classification manner (positive vs. negative), using a single public dataset which consists of product reviews. In this paper, we respond to the need for a media surveillance project to possess a custom multinomial SA classifier for usage in a restrictive and specific production setup. We describe in detail how such a classifier was built, with the help of an English dataset (containing around 15 , 000 tweets) translated to Romanian with a public translation service. We test the most popular classification methods that could be applied to SA, including standard machine learning, deep learning and BERT. As we could not find any results for multinomial sentiment classification (positive, negative and neutral) in Romanian, we set two benchmark accuracies of ≈78% using standard machine learning and ≈81% using BERT. Furthermore, we demonstrate that the automatic translation service does not downgrade the learning performance by comparing the accuracies achieved by the models trained on the original dataset with the models trained on the translated data. [ABSTRACT FROM AUTHOR]

The widespread use of social networks has brought to the foreground a very important issue, the veracity of the information circulating within them. Many natural language processing methods have been proposed in the past to assess a post's content with respect to its reliability; however, end-to-end approaches are not comparable in ability to human beings. To overcome this, in this paper, we propose the use of a more modular approach that produces indicators about a post's subjectivity and the stance provided by the replies it has received to date, letting the user decide whether (s)he trusts or does not trust the provided information. To this end, we fine-tuned state-of-the-art transformer-based language models and compared their performance with previous related work on stance detection and subjectivity analysis. Finally, we discuss the obtained results. [ABSTRACT FROM AUTHOR]

In social media, users are spreading misinformation easily and without fact checking. In principle, they do not have a malicious intent, but their sharing leads to a socially dangerous diffusion mechanism. The motivations behind this behavior have been linked to a wide variety of social and personal outcomes, but these users are not easily identified. The existing solutions show how the analysis of linguistic signals in social media posts combined with the exploration of network topologies are effective in this field. These applications have some limitations such as focusing solely on the fake news shared and not understanding the typology of the user spreading them. In this paper, we propose a computational approach to extract features from the social media posts of these users to recognize who is a fake news spreader for a given topic. Thanks to the CoAID dataset, we start the analysis with 300 K users engaged on an online micro-blogging platform; then, we enriched the dataset by extending it to a collection of more than 1 M share actions and their associated posts on the platform. The proposed approach processes a batch of Twitter posts authored by users of the CoAID dataset and turns them into a high-dimensional matrix of features, which are then exploited by a deep neural network architecture based on transformers to perform user classification. We prove the effectiveness of our work by comparing the precision, recall, and f 1 score of our model with different configurations and with a baseline classifier. We obtained an f 1 score of 0.8076, obtaining an improvement from the state-of-the-art by 4%. [ABSTRACT FROM AUTHOR]

Identifying internet spam has been a challenging problem for decades. Several solutions have succeeded to detect spam comments in social media or fraudulent emails. However, an adequate strategy for filtering messages is difficult to achieve, as these messages resemble real communications. From the Natural Language Processing (NLP) perspective, Deep Learning models are a good alternative for classifying text after being preprocessed. In particular, Long Short-Term Memory (LSTM) networks are one of the models that perform well for the binary and multi-label text classification problems. In this paper, an approach merging two different data sources, one intended for Spam in social media posts and the other for Fraud classification in emails, is presented. We designed a multi-label LSTM model and trained it on the joint datasets including text with common bigrams, extracted from each independent dataset. The experiment results show that our proposed model is capable of identifying malicious text regardless of the source. The LSTM model trained with the merged dataset outperforms the models trained independently on each dataset. [ABSTRACT FROM AUTHOR]