Sentiment analysis for user reviews using Bi-LSTM self-attention based CNN model

In this digital era, people are increasingly sharing their opinions in online review sites, and vast amounts of customer feedback are generated daily. Many customers make use of these unstructured text reviews feedback for their decision making. However, the customer feedback is exponentially huge,...

Full description

Saved in:
Bibliographic Details
Published in:Multimedia tools and applications 2022-04, Vol.81 (9), p.12405-12419
Main Authors: Bhuvaneshwari, P., Rao, A. Nagaraja, Robinson, Y. Harold, Thippeswamy, M. N.
Format: Article
Language:English
Subjects:
Artificial neural networks
Classification
Computer Communication Networks
Computer Science
Customer feedback
Customers
Data mining
Data Structures and Information Theory
Decision making
Deep learning
Electronic commerce
Engineering
Feedback
Multimedia
Multimedia Information Systems
Neural networks
Optimization techniques
Product reviews
Semantics
Sentences
Sentiment analysis
Special Purpose and Application-Based Systems
Subjectivity
Unstructured data
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this digital era, people are increasingly sharing their opinions in online review sites, and vast amounts of customer feedback are generated daily. Many customers make use of these unstructured text reviews feedback for their decision making. However, the customer feedback is exponentially huge, and reading all the reviews is tedious. Due to the maximum length of sentences, textual order changes, and logical complications, it is still a challenging area to predict the exact sentiment polarities of the user textual feedback reviews for a given entity. To address these problems, we introduce Bi-LSTM Self Attention based Convolutional Neural Network (BAC) model for subjectivity classification of reviews. In our approach, we employ pre-trained word embedding to reduce the dimension of the text representation, which avoids data sparsity issues. We also apply an attention mechanism to capture n-gram features and focus on the crucial information from the context by setting different weights between words and sentences. The proposed model uses CNN and Bi-LSTM to automatically learn classification features as well as capture the semantic and contextual information which is crucial in determining the sentiment polarities. To assess the performance of the proposed BAC model, it is compared with other baseline methods. The proposed model has achieved the accuracy of 89% and the F1-measure value of 91%.
ISSN:1380-7501
1573-7721
DOI:10.1007/s11042-022-12410-4