6.1 Introduction

To mine useful knowledge from the biomedical literature, we should be able to recognize names of biomedical entities, such as genes, proteins, cells, and diseases ^{[1] [2]} Named entity recognition (NER) refers to the task of recognizing such entity-denoting expressions, or named entities (NE), in natural language documents. In this chapter, we will be mostly concerned with biomedical NER, or NER in biomedicine.

The primary goal of NER is to relate each NE of importance in a natural language document to an individual in the real world. In this regard, NER is different from automatic term recognition (ATR), as introduced in Chapter 4, whose main goal is to associate a given term with a concept in a well-defined semantic framework. In a practical text mining system, the recognized NE then would be further classified with the known types of real-world entities. For example, if the NE denotes a gene, then it would be classified into the class of gene names. The classified NE is then linked to, or grounded by, a concept in the semantic hierarchy. This latter process for biomedical applications takes into account only the biological functions of the biomedical entity denoted by the NE.

Biomedical NER is certainly a nontrivial task, despite the availability of many well-known nomenclatures for biomedical entities, including the one published by the HGNC (http://www.gene.ucl.ac.uk/nomenclature/) for human genes ^[3] These resources do not address certain issues in NER, such as ambiguities and aliases of...