The paper presents a novel approach for complex question answering (CQA) using a Manta Ray Optimized Deep Contextualized Bidirectional Long Short-Term Memory network combined with Adaptive Galactic Swarm Optimization (MDcBiLSTM-AGSO). It addresses common CQA challenges such as query structure neglect and noisy queries by introducing a new similarity measure, 'Infoselectivity', for accurate answer classification. The proposed method outperformed existing techniques, achieving an average accuracy of 98.2% in experimental results.