Evolution of Knowledge in Social Media and Their Relationship to an Evolving Real World
Evolution of Knowledge in Social Media and Their Relationship to an Evolving Real World
复制标题
DOI：
10.1109/cogmi58952.2023.00013
复制
发表时间：
2023-11
期刊：
影响因子：
--
通讯作者：
C. Pu;Abhijit Suprem;A. Musaev;J. Ferreira
中科院分区：
文献类型：
--
作者：
C. Pu;Abhijit Suprem;A. Musaev;J. Ferreira
物质世界在进化。网络世界随着大数据的发展而发展，社交媒体是信息增长的主要组成部分。经典的机器学习模型受到其静态训练数据的限制，这些数据具有隐式的完整和永恒的知识假设。在一个不断发展的世界中，静态训练数据由于真正新颖及时的信息而遭受知识过时的困扰。知识过时导致静态机器学习模型与不断发展的世界之间的距离越来越大，称为网络物理差距。新模型的周期性再训练可能会暂时恢复其准确性，但随后它们的性能将随着网络物理差距的扩大而恶化。知识过时影响任何规模的静态训练模型，包括法学硕士。网络物理差距带来了两个主要的研究挑战：(1)收集和整合时空感知的地面真相训练数据；(2)理解和捕获物理世界和网络世界演变时信息和知识的变化速度。
The physical world evolves. The cyber world evolves and grows with big data, with social media as a major component of information growth. Classic ML models are limited by their static training data with implicit Complete and Timeless Knowledge assumptions. In an evolving world, static training data suffer from knowledge obsolescence due to truly novel timely information. Knowledge obsolescence introduces a widening distance between static ML models and the evolving world, called cyber-physical gap. Periodic retraining of new models may restore their accuracy temporarily, but subsequently their performance will deteriorate with widening cyber-physical gap. Knowledge obsolescence affects statically trained models of any size, including LLMs. Two major research challenges arise from cyber-physical gap: (1) collection and incorporation of space-time aware ground truth training data, and (2) understanding and capturing of the varying speed of information and knowledge evolution when the physical and cyber worlds evolve.