Predicting Conversations: Innovating Chatbot Technology with a Hybrid PSO-LSTM Deep Learning Model

Abstract

This research presents a novel hybrid approach combining Particle Swarm Optimization (PSO) with Bidirectional Long Short-Term Memory (Bi-LSTM) networks to enhance chatbot performance in natural language understanding and response generation. The proposed model leverages the memory capabilities of Bi-LSTM for sequential pattern recognition while utilizing PSO for hyperparameter tuning and weight optimization. By integrating these methods, the model achieves significant improvements in training accuracy (0.9961), validation accuracy (0.9854), and testing accuracy (0.9882) compared to conventional Bi-LSTM networks. The evaluation, conducted on a real-world conversational dataset, demonstrates the model's effectiveness through various metrics, including BLEU, METEOR, Word Error Rate (WER), and F1-score, establishing its robustness and reliability for intelligent chatbot systems.