Hybrid Human-Contextual Chatbot with Continuous Model Development using Recurrent Neural Network for Help-Desk Supporting Tool

Abstract

The development of chatbots is currently quite significant, considering the trend of interactive customer services 24/7. One advantage of using chatbots is reduc- ing queues and customer waiting times. However, previous research stated that 87% of users prefer interaction with humans to solve complex problems. Therefore, this study introduced a hybrid human-contextual chatbot with sustainable model devel- opment using Recurrent Neural Network (RNN) and threshold optimization. The research proposes a cooperation framework between Artificial Intelligence (AI) and humans to optimize the workforce while maintaining the quality of the company’s services. The system has a monitoring website and continuous model development to ensure the continued growth of the model. The system trial was conducted in XYZ company’s IT management division in Indonesia for four weeks. The performance evaluation process uses accuracy, hand-off rate, average execution time, and average response time. Weekly performance evaluation results obtained accuracy and hand- off rate score increased, but average execution time and response time decreased every week. The decrease in execution and response time indicates a faster model performance. The highest accuracy and hand-off rate values were 0.99 and 0.98, respectively. Execution and response times get the lowest seconds at 0.39 seconds and 0.85 seconds, respectively.