Attention Enhanced Deep Neural Network with Effective Embedding Techniques for Sentiment Analysis
Loading...
Date
item.page.authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
With the growth of the Internet and the existence of various social media, people
newlinehave opportunities to share their thoughts on topics viz., social, financial, political,
newlinereligious, entertainment, etc. User-generated reviews on products are expanding rapidly
newlinewith the emergence of e-commerce. These reviews are valuable to business organizations
newlinefor improving their products and to individual consumers for making decisions about
newlinethe entity. These online platforms consist of valuable information and an enormous
newlineamount of data for sentiment analysis. It is very difficult to handle the sentiment
newlineanalysis manually. An effective and advanced data analytic technique is required to
newlineextract intelligent opinionated content from the enormous amount of data.
newlineIdentifying and detecting opinionated content play a major role in taking an important
newlinebusiness decision. Sentiment terms do not have an explicit definition and they are
newlinewritten in user convenient language. Hence, it is difficult to identify and categorize
newlinethese contents. Influencing terms may spread across a sentence that contributes to the
newlinepolarity of the tweets. These terms must be identified, related and weighted with other
newlineterms in a sentence. The relative terms that are distributed across multiple sentences
newlineof the review, which influence the sentiment of the review are also not considered. A
newlineword may have a different meaning in various contexts and there are very few works that
newlinefocus on identifying the opinionated content in the early stages. A shallow deep learning
newlinemodel cannot categorize opinionated terms. A deep contextual model is required to find
newlinethe opinionated words from a review having multiple meanings or relations with other
newlinewords. To handle the above issues in the sentiment analysis task, we have developed
newlinedifferent architectures.
newlineHybrid Convolution Bidirectional Recurrent Neural Network (CBRNN) is introduced
newlineto overcome the issue related to identifying contextual terms. Word Embedding and
newlineSelf-Attention are combined with Long Short Term Memory (WE-SALSTM)