Article Preview
Top1. Introduction
Most of the data available on the web is not semantically structured. This results into not-precise and high-recall answers of queries by web search engines. Tools and techniques need to be developed for extraction and semantically structuring the web data, which would then enable answering meaningful and precise queries. One of the very important semantics that need to be extracted from the data is named entities (NEs). The term “named entity” means names of persons, organizations and geographical locations (Grishman & Sundheim, 1996). Named entity recognition (NER) is the process of identifying named instances of pre-defined classes in running text (Goyal, Gupta, & Kumar, 2018).
Development of NER systems has been encouraged to address the challenges in the domain, by several editions of evaluation events such as the message understanding conference, MUC (Chinchor, 1998; Sundheim & Grishman, 1996), the conference on natural language learning, CoNLL 2003 (Tjong, Sang, & Meulder, 2003), the automatic content extraction program, ACE (Nist et al., 2004) and the open knowledge extraction challenge, OKE (Rene Speck, Roder, Oramas, Espinosa-Anke, & Ngomo, 2017). The subsequent editions of these conferences, MUC, CoNLL, ACE and OKE have contributed significantly in the domain of entity detection and recognition.
NER is considered an important component of information retrieval and knowledge extraction applications, for instance, in automatic narrative extraction with multiple accounts nested in news articles (Zhang, Boons, & Batista-Navarro, 2019), text summarization (Nagalavi & Hanumanthappa, 2019), question answering systems (Pizzato, Moll, & Paris, 2006), document theme extraction (Nagrale, Khatavkar, & Kulkarni, 2019), prediction and recommendation tasks in academic citation networks (Ganguly, 2019), document indexing (Humphreys, Calcagno, & Powell, 2006), and text mining in Genetics and Biomedical Sciences (Ananiadou, Kell, & Tsujii, 2006; Settles, 2004), to name a few.
Consider the following sample text:
Salma lives in Rawalpindi and is studying Computer Science at Capital University of Science & Technology. She is a part time worker at a call center in Islamabad.
If the pre-defined classes are person, location and organization, then the output of NER task on this text is the annotated text as given below:
[person] Salma [\person] lives in [location] Rawalpindi [\location] and is studying Computer Science at [organization] Capital University of Science & Technology [\organization]. She is a part time worker at a call center in [location] Islamabad [\location].
Typically, NER demands optimally combining a variety of clues including, orthographic features, parts of speech, similarity with existing database of entities, presence of specific signature words and so on. This makes NER a non-trivial modelling challenge, not solved yet with an acceptable high precision and recall, despite over three decades of research in the field (Sarawagi, 2008). Sarawagi observes that for many existing NER systems precision is high (the entities predicted are correct), but the bigger challenge is achieving high recall (correctly predicting all the entities available in a document), because without extensive labeled data it is not possible to detect what was missed in the large mass of unstructured information.