Advanced Government Chatbot Featuring Multi-Source Strategy for Effective Goal-Driven Dialogues

Government chatbots (G-chatbots) aim to provide proactive, comprehensive information services, focusing on high-quality consultation rather than just answering queries. Due to the complexity of user questions, multiple dialogue turns are often needed for accurate responses. However, optimal answers in single turns may not align with the user's ultimate goal. To address this, we propose a Multi-Source-Aware algorithm designed for conversation-level objectives in G-chatbots. This algorithm features two key components: the Multi-Source-Aware (MSA) adapter module, which fine-tunes the chatbot's understanding based on user intent, conversation context, and preset knowledge, and the Virtual Edge module, which guides the conversation toward the final goal. Experiments show that this approach improves QA efficiency and answer quality, with fine-tuning using just 0. 01-13% of the parameters, resulting in a 2% increase in question comprehension and a 6% boost in answer quality compared to baseline models.