Development of a mobile chatbot for restaurant recommendation

Lew, Yun Cheng (2026) Development of a mobile chatbot for restaurant recommendation. Final Year Project, UTAR.

PDF Download (7Mb)

Abstract

Kuala Lumpur is home to thousands of restaurants, yet diners frequently face decision fatigue when choosing where to eat. Existing platforms rely on keyword search or popularity sorting and rarely account for individual taste history, halal certification, or dish level queries such as "chicken chop" or "ramen". This project aimed to design MakanBot, a mobile chatbot for restaurant recommendation in Kuala Lumpur that delivers personalised, halal aware, and explainable suggestions through natural language interaction. The system was built using a public dataset of 4,439 restaurants and approximately 78,000 reviews, which underwent five preprocessing steps including deduplication, text normalisation, cuisine inference, and halal label synchronisation. Three machine learning components were trained on the cleaned data, namely a TF IDF vectoriser, a content based ensemble classifier combining Support Vector Machine and Random Forest through soft voting, and an item based K Nearest Neighbours collaborative filter from the Surprise library. These models were embedded into a Python FastAPI backend connected to a native Android application, with intent analysis and explainable AI provided by the Google Gemini API and live metadata enriched through the Google Places API. The ensemble classifier achieved an accuracy of 90.63%, an F1 score of 94.32%, and an ROC AUC of 0.94, outperforming both individual models. Functional testing across 32 test cases recorded a 100% pass rate, and a pilot user study with 10 respondents reported 90% favourable satisfaction. The significance of this project lies in integrating hybrid recommendation, large language models, and culturally aware filtering on a mobile platform. Future work will focus on dataset expansion, multilingual support, and cloud deployment.

Actions (login required)