Abstract
<title>Abstract</title> <p> COVID-19-related discussions on Twitter have generated an unprecedented volume of public-opinion data, offering valuable insights into emotional reactions, crisis communication, and societal responses to the pandemic. Accurate sentiment classification of such content is challenging owing to short noisy tweets, conversational ambiguity, class imbalance, and the sensitivity of graph neural architectures to hyperparameter choices. This paper introduces <bold>BioGraph-SentiCOVID</bold> , a calibrated adaptive-context graph neural framework for COVID-19 tweet sentiment analysis. Twitter conversations are modelled as directed multi-relational graphs, where tweets form nodes and reply/root edges encode conversational structure. The main methodological contribution is <italic>Calibrated Adaptive Context Selection</italic> (C-ACS): a context-gating mechanism that combines temperature-scaled ambiguity estimation, neighbour reliability, and hop distance to decide how much thread context each tweet should exploit before GATv2 aggregation. Node representations are derived from BERTweet, while class-weighted loss and selective oversampling address sentiment imbalance. A Genetic Algorithm (GA) optimises graph depth, attention heads, hidden dimension , dropout, learning rate, weight decay, balancing ratio, and the context-gate threshold. Evaluation on COVID-Emo and supervised SemEval-2017 Task 4A adaptation uses accuracy, macro-F1, weighted-F1, per-class F1, ablation studies, and Wilcoxon signed-rank tests over <bold>n=20</bold> runs. We additionally report emotion-to-sentiment mapping sensitivity, ACS calibration diagnostics, and a lightweight GA setting for resource-constrained use. BioGraph-SentiCOVID achieves competitive results among all evaluated baselines, with a macro-F1 of <bold>0.829</bold> on COVID-Emo and <bold>0.851</bold> under the supervised SemEval-2017 adaptation protocol. C-ACS improves the fixed-depth BioGraph variant by <bold>+0.012</bold> macro-F1 while reducing noisy message passing from irrelevant distant neighbours. </p>