PS-POS Embedding Target Extraction Using CRF and BiLSTM

  • V. Roseline, Dr. Heren Chellam G.

Abstract

Sentiment analysis is one of the mostly utilized text mining applications that can trace out the sentimental view of every individual. Sentiment analysis also known as opinion mining, mines texts from website review, social network, emails, call center interactions and other information sources to identify general thread that may give positive or negative opinions on the part of customers. Deep learning uses neural networks to analyze data using iterative method that are more adaptable and intuitive than conventional machine learning tools. To improve the analysis of sentiment in a sentence effectively, we find whether the target is present in a sentence. Parts of speech (POS) is utilized for focusing the opinion expressions around the aspect word in a sentence. Using POS, a novel embedding layer called PS-POS (Predecessor Successor-Parts of Speech) is introduced in this paper. A character based Bidirectional Long Short Term Memory - Conditional Random Field (BiLSTM-CRF) was proposed to model the context information of characters. PS-POS embedding layer with BiLSTM-CRF is utilized as a sentimental analysis tool to identify the sentence of social media as single target, multi target and no target. To prove the effectiveness of our developed work, experiment was conducted on NER dataset.

Published
2020-03-30
How to Cite
V. Roseline, Dr. Heren Chellam G. (2020). PS-POS Embedding Target Extraction Using CRF and BiLSTM. International Journal of Advanced Science and Technology, 29(3), 10984 - 10995. Retrieved from https://sersc.org/journals/index.php/IJAST/article/view/27986
Section
Articles