Developing Rule-Based and AI Hybrid Chatbot for Academic Information Services

Abstract

Providing fast and accurate academic information remains a challenge in higher education, particularly when student questions are expressed informally or differ from predefined formats. Rule-based chatbots are commonly used for this purpose but often fail to recognize paraphrased or misspelled inputs, while fully generative chatbots require longer processing time and may produce unreliable responses. To address these limitations, this study developed a dual-mode academic chatbot that combined a rule-based response mechanism with a hybrid reasoning approach incorporating limited generative processing. The system was designed and refined through three iterative prototyping stages, focusing on interface usability, academic knowledge expansion, and reasoning control. After the final iteration, system performance was evaluated using ten student-generated queries that reflected real academic information needs. The evaluation showed that the rule-based mode consistently produced very fast and stable responses but achieved lower accuracy when handling non-standard inputs. The hybrid mode achieved higher response accuracy by better interpreting varied user expressions, although it required substantially longer response time. Overall, the results demonstrated that a controlled hybrid approach improved chatbot robustness while revealing clear trade-offs between response accuracy and computational efficiency.