• Ninioritse Tuedon

Student thesis: Phd


Field (2005) rightly stated that events in a movie are specifically designed to bring out the truth about the characters so that we, the audience, can transcend our ordinary lives and achieve a connection, or bond, between “them and us”; we see ourselves in them and enjoy a moment, perhaps, of recognition and understanding. The application of user ratings in making movie recommendations is implicitly detrimental to the user experience as they are deprived of the opportunity to experience movies with a storyline which they will find relatable to their present circumstances. This is because, such a movie was rated poorly by another user with similar taste to the user, or, such a movie had no ratings at all. Also, this research identified that the application of user ratings in making recommendations is also detrimental to movies with an unpopular cast list or/and movie production crew. This is because, for ratings to be predicted, initial ratings must exist, and predicted ratings are only based on existing ratings. Furthermore, the ability of a movie to have ratings is directly related to the level of popularity of the movie, because popularity enhances the visibility of the movie in the movie consumers market which in turn makes the movie available to receive a rating. If the consumers cannot see the movie, or know that the movie exists, they will not be able to provide any ratings. The application of user ratings in movie recommender systems research has created the problem of ratings-based overspecialization where movies with high ratings are seemingly handed the advantage over movies with low ratings without allowing the user to decide for themselves if they like the movie based on the movie plot. The proposed personality-based movie recommender system aims to utilize the personality of the users to identify movies suitable for the personality group using the words associated with the subject matters in movies. This involved the creation of a list of keywords based on the favourite movies provided by 207 participants from the 16 MBTI personality types. The recommendation accuracy of the proposed model based on overall user satisfaction was 76.28% when the less popular movies were recommended to the users.
Date of Award1 Aug 2021
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorPedro Sampaio (Supervisor)


  • Movies
  • Personality type
  • MBTI
  • Matrix factorization
  • word2vec
  • Recommender system
  • business model

Cite this