Matrix Factorization for Movie Recommended System Using Deep Learning

Abstract

A recommender system is a tool that provides consumers with customised material based on their prior actions. In order to develop recommender systems, this thesis investigates the influence of item and user bias in matrix factorization. User bias has been demonstrated to affect the predictability of a recommender system in previous research. To extract latent characteristics from the Movie dataset, two distinct implementations of matrix factorization using stochastic gradient descent are used, one of which takes movie and user bias into account. When it came to prediction ability, the algorithms fared equally.

There was a high association between derived characteristics and movie genres when evaluating the features retrieved from the two methods. We illustrate that each feature belongs to its own movie category, with each film representing a mix of the categories. We also demonstrate how characteristics may be utilised to suggest related films. Because human opinions assist improve product efficiency, and because a movie's success or failure is determined by its reviews, there is a growing need for and need for a good sentiment analysis model that classifies movie evaluations. Tokenization is used to convert the input text into a word vector, stemming is used to get the root of the words, feature selection is used to extract the essential terms, and classification is used to categorise reviews as positive or negative in this study. We created this model by combining KNN, SVD, and the NN.