Amalgamation of Embeddings With Model Explainability for Sentiment Analysis

Amalgamation of Embeddings With Model Explainability for Sentiment Analysis

Shila Sumol Jawale (Datta Meghe College of Engineering, India) and S.D. Sawarker (Datta Meghe College of Engineering, India)
Copyright: © 2022 |Pages: 24
DOI: 10.4018/IJAEC.315629
OnDemand PDF Download:
No Current Special Offers


Regarding the ubiquity of digitalization and electronic processing, an automated review processing system, also known as sentiment analysis, is crucial. There were many architectures and word embeddings employed for effective sentiment analysis. Deep learning is now-a-days becoming prominent for solving these problems as huge amounts of data get generated per second. In deep learning, word embedding acts as a feature representative and plays an important role. This paper proposed a novel deep learning architecture which represents hybrid embedding techniques that address polysemy, semantic and syntactic issues of a language model, along with justifying the model prediction. The model is evaluated on sentiment identification tasks, obtaining the result as F1-score 0.9254 and F1-score 0.88, for MR and Kindle dataset respectively. The proposed model outperforms many current techniques for both tasks in experiments, suggesting that combining context-free and context-dependent text representations potentially capture complementary features of word meaning. The model decisions justified with the help of visualization techniques such as t-SNE.
Article Preview


The ability to share and express feelings, feedback, and viewpoints through the Internet has grown significantly with the emergence of digitization and internet technologies Iqbal, Farkhund, et al. (2018). People’s use of social media sites like Facebook, Twitter, Instagram, WhatsApp and YouTube is on the rise. There are a variety of stakeholders who make use of these platforms, including companies and consumers as well as governments Rani, S., & Kumar, P. (2019), for a variety of reasons. Companies now have more options because of social media's increased accessibility and innovation. They employed a variety of methods to get comments and opinions from participants Jindal, K., & Aron, R. (2021). Analyzing social media content for corporate analytics, intelligence, monitoring of unethical behavior, and SA of customer opinion have all been done using a variety of methodologies Dang et al (2021). These methodologies further get more tested with interpretability/explainability through the use of explainable methods or techniques such as visualization techniques and others Jawale (2020), Van der Maaten (2008), Jawale (2023).

Since analyzing sentiment is an idea that is almost as ancient as itself, natural language processing (NLP) methods have evolved over time. Machine learning (ML) practitioners tested a wide range of user and tweet-related features, while researchers improved ML classifiers. The developments of deep learning (DL), including models from the transformer family, were also exploited, providing a plethora of current, feasible model options for each expert system creator. There is, however, always room for development. We begin with a quick survey of prior successes in the field to set the stage for our contributions towards the sentiment analysis.

An important part of NLP, known as sentiment mining, SA aims to help users analyze and recognize emotions in subjective writings. Many online communities (blogs, Twitter comments and reviews, etc.) utilize it to analyzes social media data and determine the degree of sentiment polarity in textual data. Positive, negative, or neutral polarity of an item's sentiment can be expressed in terms of how a user feels about a certain piece of text, i.e., how they feel about a given item. Online product reviews can help consumers make informed purchasing judgments based on the sentiment orientation of a wide range of customers. The SA can be classified as sentence level, document level, or word or aspect level Jawale, S., & Sawarkar, S. D. (2020, December). Document-level categorization treats the document as one unique entity, whereas sentence-level categorization considers each sentence to be its own mini-document. An aspect or word and its polarity are the only focus of an aspect-based SA .SA includes both subjectivity and objectivity into the phrase, making it a developing discipline in text mining for learning features. Sentiment analysis is done with many applications such as movie reviews Machová et al. (2020), election opinion prediction Chauhan et al. (2021), airline reviews Saad, A. I. (2020, December), and cancer hallmark classification Jiang et al. (2020), Geo-Tweets Hu et al. (2020). There are lexicon-based and machine learning/deep learning feature extraction approaches, feature extraction improves classification Drus et al. (2019). In machine learning-based approaches, the model looks for patterns in the data, whereas lexicon-based methods give lists of positive and negative terms. Each sentence consists of a combination of words. Positive and negative words determine emotion. Domain dependence makes lexicon-based approaches less suited for domains lacking specialized lexicons. Due to a lack of language resources, these strategies are typically imprecise.

Complete Article List

Search this Journal:
Volume 14: 1 Issue (2023): Forthcoming, Available for Pre-Order
Volume 13: 4 Issues (2022): 2 Released, 2 Forthcoming
Volume 12: 4 Issues (2021)
Volume 11: 4 Issues (2020)
Volume 10: 4 Issues (2019)
Volume 9: 4 Issues (2018)
Volume 8: 4 Issues (2017)
Volume 7: 4 Issues (2016)
Volume 6: 4 Issues (2015)
Volume 5: 4 Issues (2014)
Volume 4: 4 Issues (2013)
Volume 3: 4 Issues (2012)
Volume 2: 4 Issues (2011)
Volume 1: 4 Issues (2010)
View Complete Journal Contents Listing