Profiliranje spletnih uporabnikov v spletnem oglaševanju

doktorska disertacija

Domen Košir (Author), Igor Kononenko (Mentor), Zoran Bosnić (Co-mentor)

Abstract

Ogromni dobički velikih spletnih podjetij izhajajo večinoma iz naslova spletnega oglaševanja in so glavno gonilo napredka na področjih profiliranja spletnih uporabnikov ter sistemov za priporočanje. Nova metoda za gradnjo ontoloških uporabniških profilov AverageActionFC temelji na tehnikah časovnega pozabljanja in popravljanja profilov s prototipi. Prototipi predstavljajo domensko znanje, s katerim lahko občutno izboljšamo kvaliteto profila. Rezultati kažejo, da lahko z našo metodo zgradimo profile višje kakovosti kot z obstoječimi metodami. Sistemi za priporočanje, ki temeljijo na matrični faktorizaciji, trpijo za t.i. problemom hladnega zagona. Vrednosti skritih faktorjev za nove uporabnike napovedujemo na podlagi semantičnih informacij v njihovih profilih z uporabo metod strojnega učenja. Kakovost teh seznamov izdatno izboljšamo s pametnim kombiniranjem priporočil več sistemov za priporočanje.

Keywords

spletno oglaševanje;profiliranje;sistemi za priporočanje;sledenje uporabnikom;zasebnost uporabnikov;računalništvo;disertacije;

Data

Language:	Slovenian
Year of publishing:	2015
Typology:	2.08 - Doctoral Dissertation
Organization:	UL FRI - Faculty of Computer and Information Science
Publisher:	[D. Košir]
UDC:	004.7:659.1(043.3)
COBISS:	1536209091
Views:	1924
Downloads:	306
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Web User Profiling in Online Advertising
Secondary abstract:	Online advertising is a multi-billion dollar industry. Big internet companies are therefore highly motivated to improve their user profiling methods and recommendation systems. We present a novel ontological profiling method AverageActionFC. It is based on time-based forgetting and profile correction with prototypes. The prototypes are a representation of domain knowledge and can be efficiently used to improve the quality of a user's profile. The experiments show that our method significantly outperforms existing methods. Collaborative filtering recommendation systems suffer from the cold start problem. We employ machine learning algorithms to increase the quality of recommendations for new users by predicting the latent factor values based on the semantic information in their profiles. We further improve the quality of recommendation lists by combining recommendations from two or more systems
Secondary keywords:	online advertising;profiling;recommendation systems;user tracking;privacy;doctoral dissertations;theses;Oglaševanje;Disertacije;Uporabniki;
File type:	application/pdf
Type (COBISS):	Doctoral dissertation
Thesis comment:	Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages:	IX, 127 str.
ID:	8752055