Analiza in uporaba MapReduce za priporočilne sisteme

Vezočnik, Melanija (author), Jurič, Matjaž Branko (mentor)

MapReduce je programski model, namenjen za razvoj skalabilnih paralelnih aplikacij za obdelavo velikih množic podatkov, izvajalno okolje, ki podpira programski model in koordinira izvajanje programov, in implementacija programskega modela in izvajalnega okolja. Cilj diplomskega dela je analizirati MapReduce in ga preizkusiti na dveh primerih priporočilnih sistemov. Cilj smo dosegli, saj smo uspeli realizirati izračun s pomočjo MapReduce na testnih primerih. Najprej smo analizirali programski model in izvajalno okolje ter primerjali tri implementacije MapReduce: Hadoop MapReduce, MongoDB in knjižnico MapReduce-MPI. Ugotovili smo, da je za realizacijo izbranih primerov priporočilnih sistemov najprimernejša implementacija Hadoop MapReduce, saj nudi toleranco za okvare in reproducira podatke, s čimer zagotavlja zanesljivost. Nato smo z uporabo navidezne naprave Cloudera QuickStart VM, ki je gruča Hadoop z enim vozliščem, realizirali izbrana primera priporočilnih sistemov.

• Language: Slovenian
• Keywords: Hadoop MapReduce, knjižnica MapReduce-MPI, MapReduce, MongoDB, priporočilni sistemi
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2014
• PID: 20.500.12556/RUL-29534
• Publication date in RUL: 22.09.2014

Secondary language

• Language: English
• Title: Analysis and Use of MapReduce for Recommender Systems

Abstract

MapReduce is a programming model for developing scalable parallel applications for processing large data sets, an execution framework that supports the programming model and coordinates the execution of programs and an implementation of the programming model and the execution framework. The goal of the thesis is to analyse MapReduce and to use it on two examples of recommender systems. The goal is achieved by developing the computation with MapReduce successfully. At first the programming model and the execution framework are analysed and three implementations for MapReduce: Hadoop MapReduce, MongoDB and MapReduce-MPI Library are compared. It is discovered that Hadoop MapReduce is the most suitable implementation for developing the selected examples of recommender systems as it provides fault tolerance and data reproduction which ensure reliability. Then the selected examples of recommender systems are developed using Cloudera QuickStart VM which is a one node Hadoop cluster.

• Keywords: Hadoop MapReduce, MapReduce, MapReduce-MPI Library, MongoDB, recommender systems