A Conversational Movie Recommender System

Abstract

The purpose of a Conversational Recommender System is to help the users achieve their recommendation specific goals using a multi-turn dialogue. In recent years, numerous studies are conducted on improving the quality attributes of a conversational recommender system. Multiple conversational movie recommender systems are proposed. However, there is a need for a conversational system for a movie recommendation, which can be used for research purposes.

The main goal of this thesis is to create Jarvis, an open-source, rule-based conversational movie recommender system focusing on understanding the users' goals and adapting to their changing requirements. In order to understand the users' goals, a database is created, which contains the attributes with higher coverage of possible users' goals. A multi-model chat interface is designed for Jarvis. This interface introduces the components for better user interaction and providing users a guide during the conversation.

The success of a conversational system is measured in terms of the quality of the conversation and the satisfaction of the users. To guarantee the success of Jarvis, the conversation of the system with different users is recorded. Moreover, the users are requested to rate their conversation and give feedback about the system. The behavior of the system during the conversation and user feedback is studied to improve Jarvis.

The results have shown that conversational data and users' feedback plays an essential role in improving the performance of Jarvis. The users' satisfaction has improved, and the system adapts better to the previously unknown scenarios in the conversation. However, to make the system more adjustable and user-friendly, more users are required to test the system.