Back to search

Aristotle University Of Thessaloniki (AUTH)

Τεχνικές μηχανικής μάθησης για σημασιολογική δεικτοδότηση κειμένου

Abstract

dc:description

This thesis presents an original research in the field of machine learning andmore specifically in the areas of multi-label learning and topic models. Themain task considered in this thesis involves a setting in which there exists acollection of data and we are interested in determining the concepts describingeach of these data points.In the case of multi-label learning, the task is supervised: there exist atraining data set and a given concepts ontology and the goal is to train amulti-label learning model that will effectively predict concepts on new data.Topic models entail an unsupervised context, aiming at specifying both theontology of concepts, or, in other words, the topics of the collection, as wellas the concepts describing each data point. Throughout the dissertation theapplication domain is text, nevertheless the contributions of the thesis canreadily be extended to other data domains.In this thesis, we are interested in the following problems: a) effectivecombination of multi-label learning methods, b) improving parameter estimationfor supervised and unsupervised topic models c) adaptation of supervisedtopic modelling algorithms, to deal successfully with multi-label learning taskshaving extremely large concept taxonomies.First, we present a multi-label ensemble method, that combines multilabelclassifiers and employs a statistical significance test in order to validateits choices. Second, two novel estimators are introduced for topic models.These methods, employ a full probability distribution to compute improvedestimators for the topic model parameters, both in terms of performance andtime complexity. Finally, we propose an extension to multi-label learningtopic models, that effectively scales up to tasks with hundreds of thousandsof concepts. The above methods are evaluated and compared against therelevant state-of-the-art, in multiple experimental settings and for diverseevaluation procedures.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2018

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Papanikolaou, Yannis
  • Παπανικολάου, Γιάννης

Subjects

dc:subject × 6

Rights

dc:rights
Statement dc:rights
  • BY
Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/44838

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Papanikolaou, Yannis; Παπανικολάου, Γιάννης. Τεχνικές μηχανικής μάθησης για σημασιολογική δεικτοδότηση κειμένου. Aristotle University Of Thessaloniki (AUTH), 2018. http://hdl.handle.net/10442/hedi/44838