Unit:
Τομέας Υπολογιστικών Συστημάτων και ΕφαρμογώνLibrary of the School of Science
Author:
Ζεάκης Αλέξανδρος
Supervisors info:
Αναστασία Κριθαρά, Γεώργιος Παλιούρας, Παναγιώτης Σταματόπουλος
Original Title:
Author Profiling in Social Media using Topic Modeling Methods
Translated title:
Αναγνώριση Χαρακτηριστικών Συγγραφέα σε Μέσα Κοινωνικής Δικτύωσης με τη χρήση τεχνικών Θεματικής Μοντελοποίησης
Summary:
In Author Profiling Task, researchers, given a number of texts, try to find the
characteristics of the author, e.g. Age and Gender, based on stylistic- and
content based
features. In this thesis we tried to solve the Author Profiling Task utilizing
topic
modeling methods, such as Latent Semantic Indexing and, mainly, Latent Dirichlet
Allocation. To this end, we represented each document as a mixture of topics
and then
used this latent representation as input features for known classification
algorithms,
such as Support Vector Machine, to create our predictive system. To be noted,
our
approach was a part of the solution that was submitted to the 4th Author
Profiling Task at
PAN 2016.
We used two corpora for this task, one based on blogs and one on tweets, while
all
documents were preprocessed by known Natural Language Processing (NLP) methods.
The development of the system consists of phases, where in each one specific
parameters of the model were optimized and finalized. Experimental results show
that topic modeling and, in general, the proposed methodology make for good
descriptors regarding age and gender of the authors and also provides us with
new means to explore the discussion themes among age groups and genders.
Keywords:
Topic Modeling, Latent Dirichlet Allocation, Author Profiling, Information Retrieval, Latent Semantic Indexing
Number of index pages:
9, 11, 14
File:
File access is restricted only to the intranet of UoA.
document.pdf
1 MB
File access is restricted only to the intranet of UoA.
