Document-level Event Extraction via Heterogeneous Graph-based Interaction Model with a Tracker
Summary
local and global contexts are jointly modeled(局部和全局的上下文被联合建模).。
global view: Calculate the similarity between a certain phrase and the whole document in the vector space as transitional embedding based models do .(在向量空间中计算某个短语与整个文档的相似度,就像基于过渡嵌入的模型那样)
local view:
first, build a graph structure based on the document where phrases are regarded as vertices and the edges are similarities between vertices,(首先基于 document 构建一个图结构,将短语视为顶点,边为顶点之间的相似点。)
Then, proposed a new centrality computation method to capture local salient information based on the graph structure.(提出了一种基于图结构的中心性计算方法捕获局部的显著性信息)
finally, we further combine the modeling of global and local context for ranking(将全局和局部上下文的建模结合起来进行排名)
Author
Runxin Xu, Key Laboratory of Computational Linguistics, Peking University
Research Objective(s)
Entity extraction: as argument candidates
Event types detection: expressed by the document
Event records extraction: finding appropriate arguments for the expressed Events from entities.
Background / Problem Statement
Document-level event extraction aims to recognize event information from a whole piece of article.
Method(s)
For each Entity Mention node e, Averaging the representation of the contained words.
For each Sentence Node s, Maxpooling all the representation of words within the sentence plus sentence position embedding.
Sentence-Sentence Edge (S-S) : Sentence nodes are fully connected to each other with S-S edges.
Sentence-Mention Edge (S-M) : Specifically, the edge connecting the mention node and the sentence node it belongs to.
Intra-Mention-Mention Edge : connect distinct entity mentions in the same sentences(句内).
Inter-Mention-Mention Edge : (句间)
Evaluation
作者如何评估自己的方法?实验的setup是什么样的?感兴趣实验数据和结果有哪些?有没有问题或者可以借鉴的地方?
Conclusion
作者给出了哪些结论?哪些是strong conclusions, 哪些又是weak的conclusions(即作者并没有通过实验提供evidence,只在discussion中提到;或实验的数据并没有给出充分的evidence)?
Notes(optional)
不在以上列表中,但需要特别记录的笔记。
References(optional)
列出相关性高的文献,以便之后可以继续track下去。