BanFakeNews

观察到假新闻在政治和金融等各个领域的快速传播可能造成的损害，使用语言分析自动识别假新闻引起了研究界的关注。然而，这种方法主要是为英语开发的，其中低资源语言仍然不在关注点。但虚假和操纵性新闻所产生的风险并不受语言的限制。在这项工作中，我们提出了一个约 50K 新闻的注释数据集，可用于为像孟加拉语这样的低资源语言构建自动假新闻检测系统。此外，我们提供数据集分析，并使用最先进的 NLP 技术开发基准系统，以识别孟加拉假新闻。为了创建这个系统，我们探索了传统的语言特征和基于神经网络的方法。我们预计该数据集将成为构建技术以防止假新闻传播并为使用低资源语言的研究做出贡献的宝贵资源。

Given the observed harms caused by the rapid spread of fake news across various domains such as politics and finance, automatic fake news detection via linguistic analysis has attracted considerable attention from the research community. However, such approaches have primarily been developed for English, leaving low-resource languages largely unaddressed. Yet the risks posed by false and manipulative news are not constrained by language. In this work, we present an annotated dataset of approximately 50K news articles, which can be used to build automatic fake news detection systems for low-resource languages such as Bengali. Additionally, we provide a comprehensive dataset analysis and develop benchmark systems for detecting Bengali fake news using state-of-the-art NLP techniques. To build these systems, we explore both traditional linguistic features and neural network-based methods. We anticipate that this dataset will serve as a valuable resource for developing technologies to combat the spread of fake news and advance research focused on low-resource languages.

AI搜集汇总

背景与挑战

背景概述

BanFakeNews是一个孟加拉语假新闻检测数据集，包含约50K新闻的注释数据，专门针对低资源语言设计，旨在支持自动假新闻检测系统的开发。数据集由休斯顿大学和沙赫贾拉尔科技大学于2020年发布，提供了基准系统分析，探索了传统语言特征和神经网络方法，以促进孟加拉语假新闻识别技术的研究。