A High-Efficiency Government Chatbot: Multisource-Aware Algorithm Oriented Conversation-Level Goal

Abstract

Government chatbots (G-chatbots) are pivotal tools for improving public service delivery by enabling efficient and high-quality information exchange between governments and citizens. In practice, however, both efficiency and quality are undermined by persistent challenges in user understanding: chatbots often capture only short-term intentions while failing to infer long-term goals, leading to topic drift away from users’ core needs; moreover, conversations may extend into knowledge areas beyond system coverage, producing excessive dialogue turns that compromise efficiency and fragmented responses that erode quality, ultimately diminishing user patience. To address these issues, this study proposes a multisource-aware G-chatbot algorithm oriented toward conversation-level goals. The algorithm introduces two novel modules: 1) the multisource-aware (MSA) Adapter, which integrates user intentions, dialogue context, and historical knowledge to enhance semantic understanding across multiturn interactions; and 2) the virtual edge mechanism, which dynamically optimizes guidance paths by uncovering implicit relations in the knowledge graph, thereby reducing redundant exchanges and aligning responses with user objectives. Experimental results demonstrate consistent gains, including a 2% improvement in intent understanding, a 6% increase in answer quality, and a 1% reduction in dialogue turns, achieved with parameter-efficient fine-tuning (0.01%–13%). This research contributes to social intelligence and social systems modeling by offering a scalable solution for optimizing government–citizen interaction.