Named entity recognition using Bi-LSTM model with pointer cascade conditional random field for selecting high-profit products

Abstract

Named entity recognition (NER) refers to recognizing objects mentioned in texts and is considered one of the most fundamental tasks in natural language processing. The authentication of named entities is not merely a matter of extracting information independently. The rise of this sector has benefited from rapid growth, especially in the e-commerce sector; numerous reviews are published that reflect consumer sentiments on different aspects of products and services such as quality, price, and more.A critical challenge lies in improving the accuracy and robustness of NER systems to address issues such as ambiguous contexts, intricate sentence structures, and domain-specific variations. Previous works on NER usually use conventional machine learning methods. However, there is still a need to improve the accuracy of identifying entities. To accomplish this goal, this work proposes a pointer cascade conditional random field-based named entity recognition procedure. A word embedding approach is initially applied to segment the word for further processing. Word vectors are provided as input to a bidirectional LSTM (Bi-LSTM) model, which extracts features from sentence or word vectors. To improve the performance of BiLSTM, a pointer network is used to generate pointer sequences for the elements of the input array. After features are extracted, the Cascade Conditional Random Field (CCRF) layer checks tag validity by learning the correlation between tags. A Python 3.7 framework is used to implement the proposed model. According to the results of the experiments, this work achieves a high accuracy of 98.54 %.