Sistema de recomendação baseado em análise de redes

Abstract

Recommender systems are a subgroup of filtering and search algorithms sets, whose main the objective is to find or predict a user’s preferences for a given object, such as forecasting movie or tv show preferences for each user on an audio-visual media platform, for example. This work proposes a system that recommends movies, or in other words, seeks to find what the user’s most likely choices are. Films recommendation in this case, first of all, consists of finding adequate modeling for this kind of dataset to establish appropriate distance metrics and the interrelationship between their constituent entities, such as actors, directors, genre, etc. To better support the characteristics of this problem, the data structure that stands out as the first choice is the graph, and the most suitable subareas of computing for this are data engineering and data science. With this initial guideline, a recommendation system was implemented whose infrastructure uses data engineering technologies and architecture. Tools such as Wandb, Mlflow, Hydra, and Python, were used to build replicable models and version the artifacts necessary for the execution of the algorithms. Neo4j, a graph database, stores the network that best represents the collected data and has the required resources to create a complete recommendation system.