Intelligent Recommendations Based on COVID-19 Related Twitter Sentiment Analysis and Fake Tweet Detection in Apache Spark Environment

Twitter has become the main platform during COVID-19, in which people from different geographical location share their opinion. The tweets reflect the mentality of the public during COVID-19 which initiates the researchers to analyze the sentiments of the public; however, existing research works on sentiment analysis acquires limited accuracy and reliability due to a lack of contextual information and the presence of sarcastic tweets. To overcome these challenges, in this paper an Intelligent Recommendations-based Twitter Sentiment Analysis (IR-TSA) is proposed. Initially, all tweets from the apache spark environment are collected for fake tweet detection using the Light Gradient Boosting Machine (Light GBM) algorithm. After that, the location-aware topic modeling is carried out in which frequently discussed topics from different locations are modeled by constructing a Directed Acyclic Graph (DAG) structure. Then, pre-processing of tweets is done through several processes in which the context information about the tweets is preserved. The pre-processed tweets are fed into the Attention-Based Bi-directional Gated Recurrent Unit (Bi-GRU) CapsNet model for context-based sentiment classification. The CapsNet layer classifies the sentiments into five classes, namely, positive, strongly positive, neutral, negative, and strongly negative based on the scores. Finally, the generation of recommendations is carried out based on the sentiments by using the Twin delayed Deep Deterministic Policy Gradient (TD 3) algorithm and recommendations are disseminated selecting the optimal influencers using the Corona Virus Optimization Algorithm (CVOA). The proposed IR-TSA is validated in the Python 3.8.3 tool and compared with validation metrics which shows that our proposed work outperforms other existing works.