<![CDATA[This study addresses the limitations of traditional workplace canteen menu information sharing, such as noticeboards and verbal announcements, which can cause delays in shift-based settings. A WhatsApp-based canteen menu chatbot was designed, implemented, and evaluated for Matahari Department Store Sunrise Mall Mojokerto (Indonesia) using Cosine Similarity for text matching. The system integrates WhatsApp messaging via the WhatsApp API with an application server and an admin web panel built on Laravel–MySQL, enabling the management of daily menus, canteen schedules, FAQ patterns, user access, and conversation logs. User queries and stored question patterns were preprocessed (lowercasing, cleaning, tokenization), vectorized using TF–IDF, and matched using Cosine Similarity with a 0.6 acceptance threshold to route responses into three classes (Static, Menu, or Schedule) or a fallback reply. Functional verification through black-box testing confirmed the correct operation of core modules and end-to-end message handling. Similarity computation was validated against a manual example (0.816 vs. 0.8165). Robustness testing on 50 employee questions (25 standard and 25 informal/misspelled) achieved 92% accuracy (46/50); retrieval effectiveness measured as micro-F1 across the three response classes was approximately 0.95, with four queries treated as unmatched. A baseline comparison showed that Cosine Similarity was more tolerant to paraphrasing than SQL keyword matching. The case study ran from August to November 2025, and conversation logs were used during the maintenance phase to refine FAQ patterns and improve coverage. Overall, the chatbot offers a lightweight and deployable solution for routine, time-sensitive internal information services in retail workplaces.]]>
Copyrights © 2026
