Abstract

This paper investigates the possibility of automatic detection of information campaigns in the absence of a priori knowledge about the fact of their running, their goals, affected objects, and target audience. We propose a general model of information campaigns and also highlight some features of hidden information campaigns. The model is suitable for describing information campaigns both in social media and in traditional media, including those outside of the Internet. A method for detecting information campaigns, which allows the problem to be solved in automatic mode, is proposed.

To confirm the efficiency of the method, an experimental study was carried out on data collected from social media. We invited some experts in related fields to label text messages and create a test corpus. To analyze the complexity of the problem, we measured the degree of cross-expert agreement. Results of the analysis confirmed the initial hypothesis that, even for professionals, the detection of hidden information campaigns is not a trivial task. Nevertheless, using the voting method, we built a test collection to study particular information campaign features, as well as compared the proposed method with individual answers provided by the experts. Result of the experiments confirmed a high potential of the proposed approach to the problem of automatic detection of information campaigns.

中文翻译：

---

一种检测信息活动的模型和方法

摘要

本文研究了在缺乏关于信息活动运行事实、目标、受影响对象和目标受众的先验知识的情况下自动检测信息活动的可能性。我们提出了一个信息运动的一般模型，并强调了隐藏信息运动的一些特征。该模型适用于描述社交媒体和传统媒体（包括互联网以外的媒体）中的信息活动。提出了一种检测信息活动的方法，该方法允许以自动模式解决问题。

为了确认该方法的有效性，对从社交媒体收集的数据进行了实验研究。我们邀请了一些相关领域的专家对短信进行标注并创建测试语料库。为了分析问题的复杂性，我们测量了跨专家的一致程度。分析结果证实了最初的假设，即即使对于专业人士，隐藏信息活动的检测也不是一项微不足道的任务。然而，使用投票方法，我们建立了一个测试集合来研究特定的信息活动特征，并将所提出的方法与专家提供的个人答案进行比较。实验结果证实了所提出的方法在解决信息活动自动检测问题上的巨大潜力。