Named entity recognition for information extraction

Abstract

Named Entity Recognition (NER) refers to the task of examining unstructured text to extract real-world objects, peoples, terms, and more, termed ‘named’ entities, to be classified into predefined categories (Person, Location, Organization, etc.). NER serves as a crucial first step for a wide range of Natural Language Processing (NLP) applications, including machine translation, question answering, and more. In this report, we explore two paradigms for NER systems, namely RNN-based systems and Transformer-based systems. We review related work that has been done leading up to the conceptualization of these systems, and explore the implementation, characteristics, and limitations of each approach. We then compare four different approaches from RNN-based NER systems and Transformer-based systems to determine their efficacy on two semantically and structurally different datasets. From our experiments, this project determined that Transformer-based models generally perform better than the RNN-based models implemented. These results are then discussed and reasons are provided to substantiate the results observed.