Why Neural Networks are Changing How We Date Documents

There's a lot of information hidden in the dates of documents. This data can be super helpful for tasks like finding the right document or summarizing events. The problem is, not all documents have clear dates, especially those floating around the internet. Figuring out when a document was created, called "document dating, " can be tricky. It's like solving a puzzle where you need to understand how time is structured within the document.

Until now, systems for document dating mostly used specific features created by hand. They ignored the internal structures of documents. But now, there's a new way: enter NeuralDater. This uses a type of neural network called Graph Convolutional Networks (GCNs). It's like giving a robot the ability to read between the lines and understand the time structure of a document. NeuralDater can look at the way words are connected (syntax) and the order of events (time). It does this in a structured way, making it really effective. Tests on real-world data show that NeuralDater beats the best current methods by a big margin—19% more accurate, which is a 45% improvement!

Actions