Semantic FAQ Chatbot Using SBERT (Sentence-BERT) and Cosine Similarity for Academic Services

Handling repetitive inquiries in academic environments requires significant time and human resources, potentially delaying service delivery. This study developed a semantic FAQ chatbot using Sentence-BERT (SBERT) and Cosine Similarity to improve efficiency and consistency of academic information services at Universitas Jambi. The system encodes user queries into dense vector embeddings and compares them with FAQ entries using cosine similarity. A dataset of 65 frequently asked questions was collected through interviews and direct observation with students, lecturers, staff, and helpdesk officers. To evaluate semantic understanding, these entries were expanded into 130 question variations using paraphrasing. Model performance was measured with a confusion matrix and standard metrics. At a similarity threshold of 0.5, the system achieved 79.2% accuracy, 81.7% precision, 96.3% recall, and an F1-score of 88.4%. The results show that SBERT effectively identifies semantically similar questions with different wordings, handling both formal and informal Indonesian queries. High recall demonstrates that most relevant questions were successfully retrieved, while precision remains sufficient to ensure reliable responses. This study demonstrates that SBERT-based semantic matching can successfully handle Indonesian academic FAQ with diverse linguistic variations, enabling 24/7 accessibility and consistent service delivery independent of staff availability. Future work should expand the dataset to include emerging queries and conduct pilot deployment to validate operational effectiveness and user satisfaction