NLP chatbot for restaurant recommendation

Abstract

Hong Kong is a gourmet paradise with many types of restaurant from different countries. Hong Kong people like dining in quality restaurants for celebrating or gathering. Therefore, there is a high demand for dining-related services, such as restaurant databases and online booking. Generally, these services provide a website that contains restaurant searching and booking form to the users. However, different websites have different user interfaces. If the users switch to another website, they need to learn the UI operation again. Besides, it is common for people to encounter difficulty in decision-making. They have no idea to decide the restaurant under the tons of choices. A chatbot with a recommendation feature would be a better choice for the users to improve the above situation. The users can interact with the chatbot in a human language like English. They can enter a description of the desired restaurant and the chatbot can provide the suggestion according to the input. This project delivered a development process of the machine learning based chatbot application with the restaurant recommender powered by Natural Language Processing (NLP). The recommender is an ensemble model combined with three LSTM estimators trained by the restaurant review data collected from OpenRice Hong Kong. The raw data were pre-processed by a complete pipeline, including translation, cleaning, and lemmatization. The chatbot application can accept and respond in both text and voice format according to users’ needs. A cloud-based text-to-speech service was integrated into the application. The core of the application – chatbot controller, handles the message flow, which can pass the user input from the mobile interface to the recommender, transfer the predictions to the web query, and respond to the users back. In order to build an accessible environment for users with special needs, such as the elderly and disabled, serval design considerations were added to the mobile application, for example, adjustable text size. The healthy diet is more and more vital nowadays, this chatbot application supports the healthy mode that can detect unhealthy keywords and send an alert to the users.