Aspect-based Sentiment Analysis with Ontology-assisted Recommender System on Multilingual Data using Optimised Self-attention and Adaptive Deep Learning Network

In recent times of application, the Natural Language Processing (NLP) and Aspect-Based Sentiment Analysis (ABSA) seek to forecast the sentiment of polarity in several components of a document or sentence. Much present research concentrates on the relationship between aspect local context and sentiment polarity. There wasn’t enough focus on the significant deep relationships between the aspect sentiment and global context polarity. Some scholars have concluded that supervised algorithms provide promising results for ABSA. However, individually labelling information to train unsupervised systems in various domains and languages is time-consuming and expensive. Therefore, for multilingual reviews, a new ABSA model with ontology for recommendations is developed in this study. The text reviews are initially gathered from traditional online sources and then preprocessed to improve text data quality. For instance, the preprocessed data is subjected to the aspect extraction process. Then, these extracted aspects are given to the self-attention and adaptive model named SATANet for ABSA, where the guided transformer network with Dilated Deep Convolutional Network (DDCN) is used to classify the sentiments. In this SATANet, the network variables are optimised with the help of the suggested Random Position of Bonobo and Reptile Search Algorithm (RP-BRSA) to improve the recommendation performance. The final recommendation is implemented using ontology-based models, and the experimental results are validated through various heuristic algorithms and previous sentiment analysis models by considering various performance metrics.