Assessment and recommendation of neural networks and precise techniques for sentiment systems analysis

• Published in: Journal of ambient intelligence and humanized computing 2023-08, Vol.14 (8), p.11285-11299
• Main Authors: Pande, Sandeep Dwarkanath, Altahan, Baraa Riyadh, Ahammad, Shaik Hasane, Mane, Ashwini Subodhan, Inthiyaz, Syed, Smirani, Lassaad K., Hossain, Md. Amzad, Rashed, Ahmed Nabih Zaki
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial Intelligence; Artificial neural networks; Chatbots; Classification; Computational Intelligence; Data mining; Deep learning; Emotions; Engineering; Machine learning; Natural language processing; Original Research; Recurrent neural networks; Robotics and Automation; Semantic analysis; Semantics; Sentiment analysis; Social networks; Suicides & suicide attempts; Systems analysis; User Interfaces and Human Computer Interaction; Voice recognition
• ISSN: 1868-5137; 1868-5145
• DOI: 10.1007/s12652-023-04643-4

Sentiment analysis (SA) is a critical research issue in the realm of emotion. Artificial intelligence (AI) recognizes the polarity of an opinion, sentiments, and the amount of sadness communicated within user's social media positions, with subjectivity of documents or digital texts. This work presents a thorough assessment and review of current approaches and algorithms for semantic analysis. The critical contribution of this work is that it gives the most up-to-date picture of research work done in SA and recent trends in the field. It presents a profound classification of these techniques. It also focuses on challenges and emerging research areas in SA. The researchers have identified various types of emotions from user's inputs. For classifying the user feelings, they have effectively employed advanced and updated deep learning classifiers/models such as long-short-term-memory (LSTM), recurrent neural network (RNN), and convolutional neural network (CNN), capsule network (CapsNet) in their works. The glove2 pre-trained normalized phrase indices is primarily used in distinguishing emotion types. Various authors have used hyper parameter tuning to avoid overfitting and readying a better model for SA.