Local Feature Enhancement for Nested Entity Recognition Using a Convolutional Block Attention Module

Abstract

Named entity recognition involves two main types: nested named entity recognition and flat named entity recognition. The span-based approach treats nested entities and flat entities uniformly by classifying entities on a span representation. However, the span-based approach ignores the local features within the entities and the relative position features between the head and tail tokens, which affects the performance of entity recognition. To address these issues, we propose a nested entity recognition model using a convolutional block attention module and rotary position embedding for local features and relative position features enhancement. Specifically, we apply rotary position embedding to the sentence representation and capture the semantic information between the head and tail tokens using a biaffine attention mechanism. Meanwhile, the convolution module captures the local features within the entity to generate the span representation. Finally, the two parts of the representation are fused for entity classification. Extensive experiments were conducted on five widely used benchmark datasets to demonstrate the effectiveness of our proposed model.

State-of-art approach for Indian Language based on NER: Comprehensive Review

“…(d) Entity Classification: Once the named entities are identified, they are categorized or labeled into predefined classes or types, such as "PERSON," "ORGANIZATION," "LOCATION," "DATE," "MONEY," etc. For example, in the sentence "Apple Inc. is based in California," the NER system would classify "Apple Inc." as an organization and "California" as a location [48], [51], [56].…”

“…Ferrero et al [99] used a dataset for multi-CoNER [100] and created a system that distinguishes possible entity candidates while making use of external knowledge from a current knowledge base to organize them into a number of predetermined fine-grained categories, overcoming the problems caused by temporal knowledge and unidentified entities. Deng et al [56] addressed the issues with a poor generalization and single attributes which results to low recall in current approaches as they offered a unique layered entity recognition model. In this regard, it is believed that one key aspect determining entity categorization is the length feature of entities.…”

“…In order to find appropriate named entity tags for articles on Wikipedia they used Graph Knowledge Graph (GKG) [104]. THE GENIA [56] THE GENIA dataset has been taken from the biomedical literature, which helps with bio-text mining system development. The dataset includes the collection of 1999 abstracts from systems that analyze and retrieve medical literature in which three medical subject terms such as human, blood cell, and copy factor that are used to gather the abstracts.…”