Document Structure Driven Named Entity Recognition

点击翻译按钮获取中文摘要

The long-term goal of this project is developing effective and efficient tools for detecting and disambiguating named entities in legal documents. Named entities are the persons, organizations, locations, etc. that are explicitly mentioned in text using proper nouns. The task of finding entities in text and assigning them a semantic type is called Named Entity Recognition (NER), and is often performed in conjunction with pronoun resolution. Named Entity Disambiguation (NED) is the task of linking a named entity in the text to a pre-existing object in a knowledge base of interest, thus grounding the mention in something analogous to a real world entity. Often, both tasks are performed together in a pipelined fashion, in what is called Named Entity Resolution and Disambiguation (NERD). This project will accomplish two complimentary goals. The first is technology transfer: we will implement existing algorithms for Named Entity Recognition and Disambiguation (NERD) developed by the applicant and his HQP as robust and scalable Web services on Compute Canada's infrastructure. The second is to start a research collaboration that will lead to a better NERD framework, capable of handling legal jargon. A key improvement will be to make the current NERD tools use modern graph storage and computing frameworks such as Apache's Giraph, thus reducing their memory footprint. This service will be hosted by Compute Canada and will be made available to all interested researchers in Canada.