MultiCaSB-CFs: A Hybrid Explainable Deep Learning Framework for Insurance Document Generation and Query Clarification

Insurance businesses face many challenges due to outdated, slow processes that leave customers unsatisfied. Some of these challenges arise from chatbot systems’ inability to understand complex insurance-related questions, given the available cultural context and prior customer experiences. To solve these problems, we have developed an innovative architecture called MultiCaSB_CFs (Multi-Create Chatbot System Framework) that enables intelligent creation of insurance documents and provides real-time clarification for customer inquiries/tickets. The new framework uses a multi-step data preprocessing stage that includes data cleaning and Principal Components Analysis (PCA) to create features for further learning on insurance policies/tickets. The MultiCaSB_CFs incorporates: Multi-Head Cross Attention, Bidirectional Long Short-Term Memory (Bi-LSTM) Networks, and a Chaotic Sea Horse (CSH) Optimization algorithm. By combining all of these techniques, the MultiCaSB_CFs can extract rich, contextually relevant information to improve chatbot performance. Additionally, the new architecture will allow Counterfactuals (CFs) to be included as Explainable Artificial Intelligence (XAI) modules to identify the features used to select training samples for an insurance chatbot, thereby improving prediction accuracy while increasing transparency. The experimental results from the MultiCaSB_CFs. The theoretical underpinnings for MultiCaSB_CFs significantly surpass the accuracy, precision, sensitivity, and specificity of existing chatbot architectures, producing an accuracy of 0.99, precision of 0.98, sensitivity of 0.97, and specificity of 0.98. Therefore, MultiCaSB_CFs provides considerable benefits to the process of generating insurance documents as well as answering customer inquiries via chatbots.