Search Results (38)

Emotion detection in Brazilian Portuguese is less studied than in English. We benchmarked a large language model (Mistral 24B), a language-specific transformer model (BERTimbau), and the lexicon-based EmoAtlas for classifying emotions in Brazilian Portuguese text, with a focus on eight emotions derived from Plutchik’s model. Evaluation covered four corpora: 4000 stock-market tweets, 1000 news headlines, 5000 GoEmotions Reddit comments translated by LLMs, and 2000 DeepSeek-generated headlines. While BERTimbau achieved the highest average scores (accuracy 0.876, precision 0.529, and recall 0.423), an overlap with Mistral (accuracy 0.831, precision 0.522, and recall 0.539) and notable performance variability suggest there is no single top performer; however, both transformer-based models outperformed the lexicon-based EmoAtlas (accuracy 0.797) but required up to 40 times more computational resources. We also introduce a novel “emotional fingerprinting” methodology using a synthetically generated dataset to probe emotional alignment, which revealed an imperfect overlap in the emotional representations of the models. While LLMs deliver higher overall scores, EmoAtlas offers superior interpretability and efficiency, making it a cost-effective alternative. This work delivers the first quantitative benchmark for interpretable emotion detection in Brazilian Portuguese, with open datasets and code to foster research in multilingual natural language processing. Full article

Social media provides a vital arena for rare disease (RD) communities, fostering support, advocacy, and knowledge sharing. Rare Disease Day generates a large-scale online conversation, yet previous research has relied mainly on static, cross-sectional snapshots. This study captures the longitudinal evolution of the Spanish-language Twitter debate around Rare Disease Day across a fixed yearly window (1 February to 15 March) from 2008 to 2023. After filtering for Spanish-language posts, a corpus of 308,823 tweets (72,740 originals) was analyzed. We combined hashtag frequency analysis to assess topic salience with social network analysis (SNA) of co-occurrence networks to identify central thematic clusters. Results show progression from early generic expressions to increasingly deliberate, action-oriented communication, reflecting a shift towards empowered activism. A headline finding is the structural centrality and persistence of the hashtag #investigación (#research), underscoring the community’s enduring call for scientific progress. SNA further revealed the difference between transient virality—often linked to political or celebrity-driven hashtags—and the stable, identity-related topics at the core of the debate. Longitudinal hashtag analysis, particularly using SNA, provides a powerful tool to identify stable priorities of online health communities beyond transient media noise. Full article

The Olympic Games play a crucial role in shaping and promoting the host country’s national image and global perceptions. Nevertheless, limited scholarly attention has been devoted to examining how international media coverage of such events influences the perception of the host country abroad, particularly regarding major sporting events held in China. This study seeks to fill this gap by analyzing 50 China-related pieces of news from leading international publications covering the Opening and Closing Ceremonies of the 2022 Winter Olympics. Drawing from these selected news articles based on circulation metrics, this study employs a dual-level analytical framework from the perspectives of macro and micro discourses. The research integrates a corpus-assisted methodology with critical discourse analysis to systematically explore features of media headlines. We incorporate both keyword analysis and keyword-in-context approaches (KWIC) to reveal underlying patterns and meanings. Analysis of international media coverage during the Opening and Closing Ceremonies of the 2022 Beijing Winter Olympics revealed distinct narrative patterns concerning Chinese diplomatic relations and leadership. The findings indicate that foreign media outlets devoted limited attention to the Olympic events themselves. Instead, they emphasized broader sociopolitical issues, particularly in portraying China as a country that overworks regional ethnic minorities and has human rights problems. In addition, General Secretary Xi’s presidential image emerged as intrinsically linked to China’s national image in international discourse. These insights offer valuable perspectives on China’s diplomatic positioning and suggest implications for future approaches to national image construction through major sporting events. Full article

In the manufacturing supply chain, management reports often begin with concise messages that summarize key inventory insights. Traditionally, human analysts manually crafted these summary messages by sifting through complex data—a process that is both time-consuming and prone to inconsistency. In this research study, we present an AI-based system that automatically generates high-quality inventory insight summaries, referred to as “headline messages,” using real-world inventory data. The proposed system leverages lightweight natural language processing (NLP) and machine learning models to achieve accurate and efficient performance. Historical messages are first clustered using a sentence-translation MiniLM model that provides fast semantic embedding. This is used to derive key message categories and define structured input features for this purpose. Then, an explainable and low-complexity classifier trained to predict appropriate headline messages based on current inventory metrics using minimal computational resources. Through empirical experiments with real enterprise data, we demonstrate that this approach can reproduce expert-written headline messages with high accuracy while reducing report generation time from hours to minutes. This study makes three contributions. First, it introduces a lightweight approach that transforms inventory data into concise messages. Second, the proposed approach mitigates confusion by maintaining interpretability and fact-based control, and aligns wording with domain-specific terminology. Furthermore, it reports an industrial validation and deployment case study, demonstrating that the system can be integrated with enterprise data pipelines to generate large-scale weekly reports. These results demonstrate the application and technological innovation of combining small-scale language models with interpretable machine learning to provide insights. Full article

In the paper, using the three-statement financial modelling methodology as applied to a representative development project, we aim to analyse, ex ante, the industry-level impact of transition to mandatory escrow schemes in residential and mixed-use construction in Tashkent city (due to be implemented in Uzbekistan from 2026). Modelling single-milestone escrow plans against the current steep-discount advance-based system of off-plans as a baseline, the model accounts for salient institutional features of the Tashkent city development market, including land auctioning, full-cycle Value-added tax (VAT) accounting, and Tax loss carryforward provisions. It also incorporates a framework for demand-driven residual valuations for the development land element. Our findings indicate practically unchanged cashflow profitability of developers on the market in question. Around 30% p.a. in nominal Free-cashflow-to-equity based IRRs expressed in the national currency, provided that the transition to the greater use of leverage in funding unfolds as expected. The disappearance of steep off-plan discounts while the transition to escrows unfolds will be countervailed by the reliance on costly loans from escrow banks. Absent the greater use of leverage, the IRR (FCFE) profitability of the developers is expected to decline by some 5%. For the apartment buyers, this is effectively equivalent to increasing property transaction prices on the primary market in line with their headline asking amounts. Thus-generated economic surplus will be partially captured by the developers and partially passed through to escrow banks, increasing their gross profits by up to $50M, p.a. due to their new role in financing Tashkent city residential developments that are still largely equity-driven. Apart from this effect, we find only a moderate financial leverage influence on developers’ profitability due to the high-interest-rate environment prevailing in Uzbekistan. We also find a demand-driven pressure on land auction prices suggested by increasingly back-loaded alterations in project cashflow profiles. This study also purports to make a material contribution to the evolving body of literature on financial modelling of apartment and mixed-use property developments by offering a flexible three-statement modelling framework with innovative endogenised equity management features. Full article

Detecting fake news is a critical challenge in natural language processing (NLP), demanding solutions that balance accuracy, interpretability, and computational efficiency. Despite advances in NLP, systematic empirical benchmarks that directly compare both classical and deep models—across varying input richness and with careful attention to interpretability and computational tradeoffs—remain underexplored. In this study, we systematically evaluate the mathematical foundations and empirical performance of five representative models for automated fake news classification: three classical machine learning algorithms (Logistic Regression, Random Forest, and Light Gradient Boosting Machine) and two state-of-the-art deep learning architectures (A Lite Bidirectional Encoder Representations from Transformers—ALBERT and Gated Recurrent Units—GRUs). Leveraging the large-scale WELFake dataset, we conduct rigorous experiments under both headline-only and headline-plus-content input scenarios, providing a comprehensive assessment of each model’s capability to capture linguistic, contextual, and semantic cues. We analyze each model’s optimization framework, decision boundaries, and feature importance mechanisms, highlighting the empirical tradeoffs between representational capacity, generalization, and interpretability. Our results show that transformer-based models, especially ALBERT, achieve state-of-the-art performance (macro F1 up to 0.99) with rich context, while classical ensembles remain viable for constrained settings. These findings directly inform practical fake news detection. Full article

The accounting methodology for renewable energy in the European Union’s (EU) renewable heating and cooling targets is often treated as a mere technical detail, yet it has profound implications for the effectiveness of climate policies. This paper highlights a critical misalignment within the Renewable Energy Directive (RED), which inadvertently disincentivises the deployment of more efficient heating technologies. By accounting for the energy harnessed to produce the useful heat, rather than the useful heat itself, the current metrics disproportionately credit the least efficient heating systems with generating the most renewable heat. An electric heat pump with a seasonal performance factor of 3 producing 100 units of renewable heat gets credited with 100 units of heat, despite using only 33 units of input energy, whereas a wood fireplace with an efficiency of 50% gets credited with 200 units of heat. The less efficient the device, the more renewable credits it receives for producing the same amount of useful heat. This misalignment undermines decarbonisation efforts by over-crediting inefficient technologies while failing to fully recognise high-efficiency solutions like heat pumps. This paper proposes revising the RED to account for useful energy output, ensuring a more accurate reflection of technology contributions. We also propose increasing the binding heating and cooling targets of 0.8 pp/year and 1.1 pp/year so that they reflect the needed contribution of the heating and cooling sector to reach the binding headline target of 42.5% by 2030. This shift would incentivise efficiency, better align with EU climate goals, and support the transition to a low-carbon heating and cooling sector in line with the 2030 emissions reduction target. Full article

This study explores possible applications of AI technology in online journalism, given the predictions that speed and adaptation to the new medium will increase the penetration of automation in the production business. The literature shows that while the human supervision of journalistic workflow is still considered vital, the journalistic workflow is changing in nature, with the writing of micro-content being entrusted to ChatGPT-3.5 among the most visible features. This research assesses readers’ reactions to different headline styles as tested on a sample of 624 students from Timisoara, Romania, asked to evaluate the qualities of a mix of human-written vs. AI-generated headlines. The results show that AI-generated, informative headlines were perceived by more than half of the respondents as the most trustworthy and representative of the media content. Clickbait headlines, regardless of their source, were considered misleading and rated as manipulative (44.7%). In addition, 54.5% of respondents reported a decrease in trust regarding publications that frequently use clickbait techniques. A linguistic analysis was conducted to grasp the qualities of the headlines that triggered the registered responses. This study provides insights into the potential of AI-enabled tools to reshape headline writing practices in digital journalism. Full article

This study focuses on the case of news headlines in current online journalism, looking into the current possibilities opened by ChatGPT to generate such texts in an attention-grabbing manner. To assess the reaction of online readers to headlines (clickbait or click-worthy), an online survey was applied, involving Romanian students. A total of 100 original human-authored articles with clickbait headlines were extracted from a relevant Romanian database. ChatGPT was used to generate alternative headlines (one clickbait and one informative) based on the original texts. The resulting corpus of 100 headline triplets was offered to students for evaluation. More than 70% of the 600 participants in the survey preferred AI-generated headlines over the human-authored ones, indicating their experiences and behaviors in media consumption. The preferred headlines were further analyzed along lexical and grammatical characteristics, and stylistically, to pinpoint the features sparking readers’ curiosity and engagement. While on a cognitive level the investigated audience rejected clickbait headlines as being deceitful and frustrating, in practice less than 34% favored neutral and objective headlines. Also, the linguistic analysis provided insights into the mechanics of reader engagement and the effectiveness of various headline strategies. The results are useful to anticipate the adoption of AI as a creative partner in Romanian media practice. Full article

Critical thinking is essential when navigating, evaluating, and interacting with media; therefore, it is important to investigate if adults’ critical thinking skills can be trained. This paper describes an experiment investigating the impact of video lectures about enthymemes and critical thinking skills on participants’ (N = 176) critical thinking skills, measured by the Watson–Glaser Critical Thinking Appraisal (WGCTA) and on their ability to identify clickbait headlines. Participants were adults recruited through the Prolific Platform, and they were randomly assigned to one of three conditions: an enthymeme lecture, a general critical thinking lecture, or a control condition. The results indicated no significant improvement in critical thinking scores across the conditions, as measured by the WGCTA. Similarly, no significant differences were found in the participants’ ability to identify clickbait headlines. However, a significant positive correlation was observed between higher critical thinking scores and better clickbait recognition. These results suggest that a short lecture-based intervention may not be sufficient to significantly improve adult learners’ critical thinking. Perhaps this study indicates the need for more in-depth or interactive interventions to effectively support media literacy. The material presented here is a kind of counterexample of what should be done. For this reason, it may prove useful in future research to avoid certain experimental dead-ends. Full article

Sentiment analysis is a well-known task that has been used to analyse customer feedback reviews and media headlines to detect the sentimental personality or polarisation of a given text. With the growth of social media and other online platforms, like Twitter (now branded as X), Facebook, blogs, and others, it has been used in the investment community to monitor customer feedback, reviews, and news headlines about financial institutions’ products and services to ensure business success and prioritise aspects of customer relationship management. Supervised learning algorithms have been popularly employed for this task, but the performance of these models has been compromised due to the brevity of the content and the presence of idiomatic expressions, sound imitations, and abbreviations. Additionally, the pre-training of a larger language model (PTLM) struggles to capture bidirectional contextual knowledge learnt through word dependency because the sentence-level representation fails to take broad features into account. We develop a novel structure called language feature extraction and adaptation for reviews (LFEAR), an advanced natural language model that amalgamates retrieval-augmented generation (RAG) with a conversation format for an auto-regressive fine-tuning model (ARFT). This helps to overcome the limitations of lexicon-based tools and the reliance on pre-defined sentiment lexicons, which may not fully capture the range of sentiments in natural language and address questions on various topics and tasks. LFEAR is fine-tuned on Hellopeter reviews that incorporate industry-specific contextual information retrieval to show resilience and flexibility for various tasks, including analysing sentiments in reviews of restaurants, movies, politics, and financial products. The proposed model achieved an average precision score of 98.45%, answer correctness of 93.85%, and context precision of 97.69% based on Retrieval-Augmented Generation Assessment (RAGAS) metrics. The LFEAR model is effective in conducting sentiment analysis across various domains due to its adaptability and scalable inference mechanism. It considers unique language characteristics and patterns in specific domains to ensure accurate sentiment annotation. This is particularly beneficial for individuals in the financial sector, such as investors and institutions, including those listed on the Johannesburg Stock Exchange (JSE), which is the primary stock exchange in South Africa and plays a significant role in the country’s financial market. Future initiatives will focus on incorporating a wider range of data sources and improving the system’s ability to express nuanced sentiments effectively, enhancing its usefulness in diverse real-world scenarios. Full article

Background: In recent years, there has been growing interest in the evaluation of autism spectrum disorder (ASD) and autistic traits in prison populations and offenders. Due to misleading headlines and highly publicized criminal cases, the belief that autistic individuals are more prone to commit crimes has spread among the general population, also leading to increasing research on this matter. Aims: In this context, this narrative review aimed to analyze the available scientific literature on the bi-directional link between ASD and criminal behaviors and to assess the key characteristics of eventual ASD offenders, including sociodemographic data, comorbidities, crime-related features, and interactions with the criminal justice system. Results: Our review highlighted that the available studies lack methodological rigor and present controversial results. Overall, the current state of research does not support any definitive correlation between ASD or autistic traits and the predisposition to engage in criminal conduct. Further studies are needed to confirm or reject this hypothesis. Full article

Headline generation aims to condense key information from an article or a document into a concise one-sentence summary. The Transformer structure is in general effective for such tasks, yet it suffers from a dramatic increase in training time and GPU consumption as the input text length grows. To address this problem, a hybrid attention mechanism is proposed. Both local and global semantic information among words are modeled in a way that significantly improves training efficiency, especially for long text. Effectiveness is not sacrificed; in fact, fluency and semantic coherence of the generated headlines are enhanced. Experimental results on an open benchmark dataset show that, compared to the baseline model’s best performance, the proposed model obtains a 14.7%, 16.7%, 14.4% and 9.1% increase in the F1 values of the ROUGE-1, the ROUGE-2, the ROUGE-L and the ROUGE-WE metrics, respectively. The semantic coherence of the generated text is also improved, as shown by a 2.8% improvement in the BERTScore’s F1 value. These results show that the effectiveness of the proposed headline generation model with the hybrid attention mechanism is also improved. The hybrid attention mechanism could provide references for relevant text generation tasks. Full article

This study examines the process of reading polycode advertising posters, focusing in particular on the effect of a pun in the headline. The pun, or a sequence of lexical items that can be perceived as ambiguous, is contained in the headline and different meanings of this sequence are supported by the picture and text. The results of the preliminary experiment showed that advertisements with puns are rated as more attractive, original, effective and positive compared to advertisements without puns. We hypothesized that puns in the headlines increase cognitive effort in processing posters, leading to higher evaluations. The main experiment tested this and examined differences in eye movement when reading posters with and without puns. Fifty-five Russian participants viewed advertisements while their eye movements were recorded. Our results showed no fundamental differences in the general pattern of viewing advertisement posters with and without puns. We found that readers start to perceive polycode advertisements from the text and spend more time reading the text than looking at an image. These findings shed light on how attention is distributed between verbal and non-verbal components of polycode texts, and which type of poster is more effective for information retrieval at different processing levels. Full article

Previous research highlighted that a fundamental rethink of the measurement of domestic abuse was needed in the Crime Survey for England and Wales (CSEW). The research reported here aimed to develop and test new questions on domestic abuse for the CSEW to improve the headline prevalence measure, including frequency of abuse, to develop a way of measuring controlling or coercive behavior within the overall prevalence measure, and to develop a measure of the impact of abuse. The research included focus groups and interviews with victims (n = 27) to assess a set of draft questions and cognitive testing of revised questions with victims and the general public (n = 42). A final set of 24 questions was developed for use with victims of both intimate partner and family abuse, with an additional question for family abuse. The new questions were found to echo victim experiences and were deemed acceptable and reliable measures by victims and the general public for domestic abuse, including controlling and/or coercive behavior and impact. An analytical approach was recommended to improve the headline prevalence measure of domestic abuse by establishing ‘high’ and ‘low’ abuse profiles using measures of both behavior and impact. Full article