ANALYSIS OF POPULARITY BIAS EFFECT REDUCTION IN RESTRICTED BOLTZMANN MACHINE RECOMMENDER SYSTEM USING BINARY XQUAD METHOD

• Main Author: Alemmario, Reyhan
• Format: Final Project
• Language: Indonesia
• Subjects: Fisika
• Online Access: https://digilib.itb.ac.id/gdl/view/47450
• Institution: Institut Teknologi Bandung

Statistical physics is a part of physics that inspires many other fields, for example the MaxwellBoltzmann distribution which can inspire solutions to problems in the field of computing with graph models. Restricted Boltzmann Machine (RBM) is one of the results that can handle a variety of problems, one of the problems is the recommender system using machine learning. The resulting recommendation system is a type of collaborative filtering which has a weakness in popularity bias which is a common problem often faced by machine learning when the distribution of objects is non-normal. To detect bias in the recommendation system a correlation of popularity has been sought with the number of ratings to obtain a biased feature. In addition, the recommendation results produced by the system have been evaluated so that we have a non-normal distribution. This problem has been dealt by the Binary Explosion Query Aspect Diversification method, which reorder the results of existing recommendations but with the aim of reducing the effect of popularity bias by weighting the results of recommendations based on the type of area occupied by the object in binary division of the area according to the method. Weighting method which is generally done before the process of creating machine learning models can be done after the process of creating model with this method