基于大模型的事件抽取技术及军事应用思考

Reflections on large model event extraction technology and military applications

事件抽取旨在从非结构化文本中抽取出结构化事件信息,以便清晰、方便、直观地掌握并利用相关的关键信息。传统机器学习方法依赖于特征工程,利用人工构建的特征来进行事件抽取。而基于深度学习的方法利用CNN、RNN、GNN等深层神经网络通过提取重要特征来展开,但其依赖于大量的标注数据。近年来,研究者开始利用基于Transformer架构的大规模语言模型如BERT、GPT等采用预训练+微调范式来进行事件抽取并取得显著成效。而最近推出的大模型ChatGPT采用预训练+提示学习范式在自然语言处理领域取得显著成效,可以实现高效准确地抽取出关键的事件信息,将其应用到军事领域会产生重大影响。

Event extraction aims to extract structured event information from unstructured text,in order to grasp and use relevant key information clearly,conveniently and intuitively.Traditional machine learning methods rely on feature engineering,using artificially constructed features for event extraction.The method based on deep learning uses CNN,RNN,GNN and other deep neural networks to extract important features,but it relies on a large number of annotated data.In recent years,researchers begin to use transformer architecture based large-scale language models such as BERT and GPT to use pre-training and fine-tuning paradigm for event extraction and achieved remarkable results.The recently launched large model ChatGPT adopts the pre-training prompt learning paradigm to achieve remarkable results in the field of natural language processing,which can realize the efficient and accurate extraction of key event information,and its application to the military field will have a significant impact.