Analyzing Fake Sports News Detection Methods Using Attention Mechanism and Neural Networks in the Sports Industry

Abstract

In an era marked by the rapid proliferation of Internet-based social platforms, efficient data dissemination across networks has reached unprecedented levels. However, this widespread accessibility to information has also ushered in a concerning trend—the propagation of false and deceptive narratives. The consequences of fake news, especially in the context of the sports sector, are far-reaching. Beyond eroding trust in media sources, fake sports news can instigate social unrest and disruption with implications extending into politics and the economy.Recognizing the urgent need to address this challenge, this study embarks on the vital task of automatically detecting fake sports news within the complex landscape of online content. Fake sports news typically comprises statements or reports containing elements of falsehood, often diverging significantly from the actual events they purport to describe. Such misinformation is frequently disseminated for political or economic reasons, making its accurate identification formidable.To tackle this pressing issue, we propose a comprehensive fake sports news detection framework termed MFNDF (Multimodal Fake News Detection in Sports). This innovative framework leverages a multifaceted approach, extracting features from three distinct modalities.Firstly, we employ the BERT model for text feature extraction, fine-tuning the extracted text features through a fully connected (FC) layer to enhance the representation of news semantics.Secondly, in image feature extraction, we harness the power of the DenseNet pre-training model to extract convolution features from image content. Additionally, we utilize the Discrete Cosine Transform (DCT) algorithm to extract image frequency domain features, which aid in detecting image tampering and repeated compression—a common tactic employed in creating fake sports news images.Thirdly, we delve into the realm of user context, mining valuable insights from users' behavioral features and news statistical features through advanced feature engineering techniques.Furthermore, we introduce an attention mechanism, a critical component of our approach, to assign weights to word vectors within the text feature space. This attention mechanism takes into account both image features and word vectors, thereby facilitating the fusion of image and text information. The resulting feature vector, enriched with multimodal data, enhances the overall performance of our detection model.Systematic experiments conducted on MFNDF confirm its superior effectiveness in the realm of fake sports news detection. By addressing the critical challenge of identifying deceptive narratives within the sports sector, our research strives to mitigate the adverse impact of fake sports news and safeguard the integrity of information in the dynamic world of sports reporting."