Abstract
Over the past few years, much work has been done to develop machine learning models that perform Arabic sentiment analysis (ASA) tasks at various levels and in different domains. However, most of this work has been based on shallow machine learning, with little attention given to deep learning approaches. Furthermore, the deep learning models used for ASA have been based on noncontextualized embedding schemes that negatively impact model performances. This article proposes a novel deep learning-based multilevel parallel attention neural (MPAN) model that uses a simple positioning binary embedding scheme (PBES) to simultaneously compute contextualized embeddings at the character, word, and sentence levels. The MPAN model then computes multilevel attention vectors and concatenates them at the output level to produce competitive accuracies. Specifically, the MPAN model produces state-of-the-art results that outperform all established ASA baselines using 34 publicly available ASA datasets. The proposed model is further shown to produce new state-of-the-art accuracies for two multidomain collections: 95.61% for a binary classification collection and 94.25% for a tertiary classification collection. Finally, the performance of the MPAN model is further validated using the public IMDB movie review dataset, on which it produces an accuracy of 96.13%, placing it in second position on the global IMDB leaderboard.
Highlights
Sentiment analysis is an important subarea of natural language processing (NLP) where advanced statistical and machine learning models are used to measure sentiments, emotions and opinions in many domains, including customer satisfaction, product acceptance, market directions, and public approval of political decisions and events
In an attempt to address the limitations of existing approaches, this article introduces a novel deep learning-based multilevel parallel attention neural (MPAN) model comprising multiple parallel channels, where each channel represents a different neural model that works in parallel with other models to perform training or testing
The MPAN model is shown to achieve an accuracy of 95.61% for binary classification collection and 94.25% for tertiary classification collection, all of which are well above the current baselines reported in the literature. (iv) the performance of the MPAN model is further validated using the public IMDB movie review dataset, producing an accuracy of 96.13%, placing it in second position on the global IMDB leaderboard above Bidirectional Encoder Representations from Transformers (BERT) [9]
Summary
Sentiment analysis is an important subarea of natural language processing (NLP) where advanced statistical and machine learning models are used to measure sentiments, emotions and opinions in many domains, including customer satisfaction, product acceptance, market directions, and public approval of political decisions and events. El-Affendi et al.: Novel Deep Learning-Based MPAN Model for Multidomain Arabic Sentiment Analysis combinations of both. These sequential models lack the flexibility needed to build arbitrary networks. Such approaches are not fully ‘‘contextualized’’ and ignore the position and order of words, negatively impacting the resulting accuracy To address these problems, some researchers have designed attention-based models that search for and amplify relevant context areas in input vectors [6]–[9]. In an attempt to address the limitations of existing approaches, this article introduces a novel deep learning-based multilevel parallel attention neural (MPAN) model comprising multiple parallel channels, where each channel represents a different neural model that works in parallel with other models to perform training or testing.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
