Objektiivne raamatute soovitussüsteem
Date
2015
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Antud töö eesmärgiks oli objektiivselt mõõdetavatel atribuutidel põhineva raamatute soovitussüsteemi loomine. Loodud süsteem kasutab soovituste andmiseks raamatute hinnan-gute põhjal õpitud lineaarse regressiooni mudelit. Töös hinnatakse antud lähenemise efektiivsust taolise probleemi lahendamisel. Mudeli põhjal on võimalik arvutada erinevusi ingliskeelsete raamatute vahel saades sisendiks analüüsitud raamatupaarid. Iga raamatu sisu puhul analüüsitakse teda kirjeldavaid atribuute nagu teksti positiivsus või keskmine lausete pikkus. Treeningandmetena kasutati andmestikku lugejate hinnangutest raamatutele ja avatud raamatute kogumiku. Tulemusena saadi mudel, mis suudab korrektselt leida erinevusi raamatute vahel treeningandmete piires, kuid ei suuda korrektselt leida inimlikust vaatepunktist sarnaseid raamatuid treeningandmete vähesuse ja kvaliteedivaesuse tõttu. Mudeli töö korrektsuse põhjal järeldati, et suurema ja parema treeningandmestiku puhul antud töös kasutatud lähenemisel on potentsiaali arendada ka inimliku subjektiivsuse vaatepunktist korrektse raamatusoovitaja.
This work’s goal was the creation of a book recommender system based on objectively measurable attributes. For recommendations, the created system uses linear regression model trained on the books’ ratings. In this work effectiveness of the described approach as a solution for the given problem is measured. The model is able to estimate the differences between two books in English. For every book’s content, a set of attributes describing the book are calculated, such as text positivity or average sentence length. Books’ ratings and open books collection were used as the training dataset. As a result, the model is able to find differences correctly in terms of the training data provided, but is not able to correctly find the nearest books from human point of view due to the lack of a large good quality dataset. Based on the model’s work correctness, it was concluded that given a big enough good quality training dataset the approach used in this work has potential to find a model capable of finding the nearest books in human understanding.
This work’s goal was the creation of a book recommender system based on objectively measurable attributes. For recommendations, the created system uses linear regression model trained on the books’ ratings. In this work effectiveness of the described approach as a solution for the given problem is measured. The model is able to estimate the differences between two books in English. For every book’s content, a set of attributes describing the book are calculated, such as text positivity or average sentence length. Books’ ratings and open books collection were used as the training dataset. As a result, the model is able to find differences correctly in terms of the training data provided, but is not able to correctly find the nearest books from human point of view due to the lack of a large good quality dataset. Based on the model’s work correctness, it was concluded that given a big enough good quality training dataset the approach used in this work has potential to find a model capable of finding the nearest books in human understanding.